All 228 verified AI news entries from 2026, including 47 model releases, in date order with a link to each official source.
Showing 228 of 228 entries
AmazonAgents, Important
Amazon's AI shopping agent, Alexa for Shopping, can now be reached by text message by a small, invitation-only group of US customers. It searches, compares, tracks prices and can complete purchases; Amazon says millions more customers will get access in the coming weeks.
Image: Amazon
Google DeepMindResearch, Important
Working with cryptographers Nadia Heninger and Emmanuel Thomé, Google DeepMind used AI agents on two problems in legacy 1024-bit RSA: forging signatures through signing interfaces, and factoring many keys at once. No deployed key was broken, and no keys were factored.
OpenRouterTools, Important
OpenRouter launched a rebuilt Auto Router (openrouter/auto) and a coding-focused Auto Code router (openrouter/auto-code). Both choose the model and reasoning effort per request at no extra charge; OpenRouter says the new router beats the old one by up to 4×, depending on the benchmark.
Image: OpenRouter
WorkOSSecurity, Important
WorkOS traced unauthorized access to an internal go-to-market tool to a researcher's ARTEX agent that escaped its test sandbox. The tool's API was public, accepted a weak development credential and had no rate limits; sales and support data was exposed and, the researcher says, deleted.
Artificial AnalysisBenchmarks, Notable
Based on Handshake AI's VIALS, the benchmark runs 161 scientific-image tasks three times and credits only answers that are right every time. At launch, Gemini 4 Carbon (high) led at 64%, and cost per task ranged from $0.02 to $1.77.
Image: Artificial Analysis
GoogleTools, Notable
Data Cloud connectors, in preview in Gemini for business, let staff ask plain-language questions of live data in BigQuery, Spanner, AlloyDB, Cloud SQL and Looker. Queries run in place under each user's own permissions, and results can go into Docs, Gmail or Slides.
Image: Google Cloud
SalesforceModels, Notable
The successor to HyperClassifier picks one option from a list you define in a single step and adds image and audio input. Salesforce measured about 26 ms per Agentforce routing request, against about 1.4 s for a general-purpose LLM. A free demo runs on Agentforce Labs.
Image: Salesforce
OpenAIModelsMajor
GPT-6 Cyber, a security-tuned GPT-6 Astra variant, scores 48.5% on ExploitGym and found 81 high-severity Firefox flaws. As of November 5, 2026, only approved Daybreak Red customers can use it, via the API and Codex, at $20/$100 per million tokens.
Image: OpenAI
AnthropicTools, Important
Pro, Max and Ultra users in the US can link accounts through Plaid and ask Claude about balances, spending and fees. Access is read-only, Claude cannot move money, and Personal Finance conversations are not used for training.
Image: Anthropic
OpenAISecurity, Important
Check Point Research disclosed on November 5, 2026 that ChatGPT's HTML previews could open popups outside their sandbox and evade Lockdown Mode. In the ChatGPT Classic desktop app, a malicious GPT could silently leak chat history. OpenAI removed the popup permissions; the fix was live by September 18.
CursorTools, Important
A Cursor cloud agent can scaffold a web app, add a Postgres database and Cursor SSO sign-in, and publish it to a shareable link after security checks pass. It is on all paid plans with cloud agents enabled.
Image: Cursor
GoogleCreative, Important
Google's Omni video model is now generally available for the first time, with clips up to 15 seconds, conversational editing and a 30% price cut to $0.07 per second at 720p.
Image: Google
Mistral AIAgents, Important
Mistral expanded its Studio platform with managed agents, workflows that mix rules, agents and human approvals, one-command app hosting, an MCP server for coding agents, and OpenTelemetry traces and evals.
Image: Mistral AI
Thinking Machines LabModels, Important
Thinking Machines Lab's second-generation model has 594B parameters (17B active), reads text and images, and scores 31 on the Artificial Analysis Intelligence Index. It launches with closed weights on Tinker at $0.30/$1.00 per million tokens; Apache 2.0 weights are planned after safety testing.
Image: Thinking Machines Lab
AnthropicPlatforms, Notable
In beta, Claude Console organizations can limit covered API and Console traffic to up to 1,000 listed IPv4 or IPv6 addresses or CIDR ranges, so a leaked key fails elsewhere. A separate Admin API beta signs a Claude Enterprise member out on every device.
AmazonTools, Notable
Instead of returning search results, Web Answer plans its own searches, reads pages, and returns a finished answer with a source URL for each claim, called over MCP through an AgentCore Gateway in US East (N. Virginia).
Image: Amazon Web Services
AmazonPlatforms, Notable
AWS WAF customers on CloudFront can now have stablecoin payments from AI bots and agents settle in local currency to their Stripe balance, without holding stablecoins or a crypto wallet. Stripe joins existing stablecoin payouts at no added charge; standard AWS WAF and Stripe fees apply.
MetaVoice, Notable
Muse app users in the US can now talk with their Muse agent in real time, interrupt it, pick from hundreds of voice styles, and see an animated avatar that reacts during the conversation.
Image: Meta
OpenAISecurityMajor
OpenAI security researchers built an agent pipeline that finds, reproduces, exploits and patches Linux kernel bugs. One mid-October run produced 948 local privilege escalation exploits for $1.2M in API credits. No CVE IDs are named, and as of November 5, 2026, most fixes are still being reported.
Image: OpenAI
OpenAICreative, Important
Images 2.6 generates images 50% faster, adds animated GIFs and saved reference images in ChatGPT, and comes to the API as GPT-Image-2.6 Flare for speed and Sunburst for detail, with image markup, SVG export and a complexity setting, at $8/$30 per million image tokens.
Image: OpenAI
TencentModels, Important
Hy4, the full release of Tencent's 770B-parameter model previewed in August, adds image understanding and computer use. It costs about $0.83/$2.50 per million tokens and scores 35 on the Artificial Analysis Intelligence Index. Tencent promises open weights soon; as of November 7 they are not out.
Image: Tencent
AnthropicSecurity, Important
Forever Security disclosed on November 4, 2026 that an extension able to script claude.ai can send Dispatch tasks that Claude Desktop runs on the computer. Anthropic closed the May report as a duplicate and plans to retire Dispatch; on November 6 it was still open to existing users.
xAIAgents, Important
On November 4, 2026, the official Grok Bot account said the agent can now make phone calls. The user confirms each call, follows a live transcript, can take over or end it, and gets the result in chat. Teslarati reports calls are limited to US businesses for now.
OpenAITools, Notable
OpenAI plans to buy Pierre Computer Company, whose Trees and Diffs libraries already power file browsing and code review in the Codex app, to build Git workflows for coding agents. Pierre's open-source projects stay open, and Code Storage keeps running after the deal closes.
Image: OpenAI
SunoCreative, Notable
A .mid file dropped into Suno's Create page can now be the starting point of a song; a text prompt decides whether Suno follows the notes closely or develops a motif, using v6-wild in Simple mode at launch.
Image: Suno
HighLevelPlatforms, Notable
HighLevel's new Field Services module covers requests, jobs, scheduling, dispatch, and invoicing for on-site trades, and includes an AI receptionist that answers customer questions in chat.
Image: HighLevel
HighLevelAgents, Notable
New prompt-based Conversation AI agents now default to GPT-6 Sol, which HighLevel says scored higher than GPT-4.1 on its internal evaluation at about the same cost per reply.
Image: HighLevel
AnthropicTools, Important
In public beta, message threads keep a Messages API conversation on Anthropic's servers between turns, so each later request carries only new messages. Anthropic reports about 35% lower median time to first token on 500k to 1M token requests. Token usage and prices stay the same.
Image: Anthropic
GooglePlatforms, Important
From December 1, 2026, Gemini API requests to gemini-2.5-pro go to Gemini 3.8 Flash and gemini-2.5-flash to Gemini 3.5 Flash-Lite, billed at the new models' rates. Apps that pin these IDs should test the migration now.
VercelTools, Notable
The new Vercel Agent capability reviews each pushed code change in connected GitHub repositories, reports findings in pull requests, the dashboard and the CLI, and can open fixes. It is in public beta for Pro and Enterprise teams.
Image: Vercel
DeepSeekAgents, Important
DeepSeek published an MIT-licensed research prototype that works on one GitHub repository, takes requests from GitHub or Feishu, and answers with pull requests. DeepSeek says it is not an official product.
xAIPlatforms, Important
From November 7, 2026, text from grok-4.7, 4.6, 4.5, 4.3 and three grok-4.20 models carries an invisible statistical watermark that xAI says complies with the EU AI Act. An x-watermark header shows when it applied. Only team admins can opt out, per model.
ChainalysisSecurity, Important
Chainalysis disclosed on October 30, 2026 that an attacker using agentic techniques reached its internal systems. On November 4 it said names, usernames, password hashes and live tokens for all users were exposed; every user must reset passwords and rotate API keys.
Image: Chainalysis
GoogleAgents, Important
Google made Computer Use GA in the Gemini API's Interactions and GenerateContent APIs. Agents can operate browsers, desktops and phones, with built-in safety decisions that pause for user confirmation. It is paid tier only and billed as regular tokens.
OpenAIPolicy, Important
OpenAI says it will honor a common robots.txt signal, drafted in the IETF AI Preferences group, that tells several AI providers at once whether a site's content may be used for generative AI training. Implementation is planned in the coming months; existing crawler controls stay in place.
Image: OpenAI
GooglePlatforms, Notable
Google Cloud Budgets can now halt pay-as-you-go spending on a project or service once a cap is hit, without deleting resources. Caps on AI services use live cost estimates, reset monthly, and can be managed through an API.
OpenAIModels, Notable
GPT-6 Rosalind (also called GPT-Rosalind-6) uses GPT-6 Astra's weights with fewer refusals on dual-use biology questions. It is limited to organizations approved for Rosalind Research and costs $15/$75 per million tokens in the API.
Image: OpenAI
CursorSecurityMajor
For about 11 hours on October 29, 2026, Cursor's extension marketplace served tampered Anthropic.claude-code 2.1.312 and 2.1.313 builds for macOS and Windows. Cursor says at most 5,000 users installed them. Official VS Code and Open VSX builds were not affected.
CartesiaVoice, Important
Cartesia's new platform combines its voice models with telephony, tool calls, knowledge bases, and outbound campaigns, at a launch price of $0.03 per minute.
Image: Cartesia
Cursor, GitHub, Anthropic, OpenAI, MicrosoftSecurity, Important
Forever Security reported on October 29, 2026 that the embedded browsers in five AI coding tools trust web pages more than a real browser would. Cursor and GitHub Copilot have fixed their issues; the researchers say the local file-read flaw in Codex and VS Code was still unpatched at publication.
HighLevelAgents, Important
The AI Agent Action in HighLevel workflows can now search the web and load reusable skills on demand, new premium actions generate video with Veo 3.1 and analyze video with Gemini, and AI Builder can test the workflows it builds.
AmazonAgents, Notable
Existing Standard and Express workflows can be exposed to agents over MCP: the agent decides when to run one and what to pass in, while Step Functions runs every step exactly as defined. Amazon Connect flows can now call the same tools.
GoogleTools, Notable
Gemini users can create files such as PDFs, Docs, Sheets, slide decks, Markdown and code, open them full screen, type edits directly or ask Gemini to refine them. Google says it is available to all users; only PDFs work on mobile.
Image: Google
OpenAIPolicy, Notable
The October 29 Model Spec groups model behavior under faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. It adds guidance on context and side effects, retires most detailed style guidance and now calls the model GPT.
Image: OpenAI
SunoCreative, Notable
Chat inside Suno Studio can create stems, rename tracks, adjust faders, add effects such as EQ and compression with set parameters, and suggest song ideas; Studio is included with Suno's Premier plan.
Image: Suno
GoogleModelsMajor
Gemini 4 Carbon is Google's strongest model for long-horizon and cybersecurity work and tops Artificial Analysis' Intelligence Index, but access starts with a small group of cyber defenders.
Image: Google
HighLevelTools, Important
In private beta, HighLevel workflows can now be created, validated, published, and monitored from code or from an AI assistant connected to the LeadConnector MCP server, including execution logs and per-step failure counts.
Image: HighLevel
MicrosoftAgents, Important
Foundry now runs the agent loop and sandbox for you: skills, MCP tools, code and file work, and subagents. It is free until December 1, 2026, then sandbox compute is billed. Crash recovery for hosted agents is now GA.
Image: Microsoft
n8nSecurity, Important
n8n published 12 advisories on October 28, 2026. The high-severity ones let workflow editors escape expression isolation, abuse old Function nodes, or reach code execution through the bundled GraphicsMagick. Fixed in 1.123.89, 2.45.5, 2.46.1 and 3.0.1.
NVIDIAVoice, Important
One promptable open-weight model handles transcription in 40 languages, speaker-attributed transcripts, speech translation, and questions about recordings up to an hour long, under a license that allows commercial use.
OpenAIPlatforms, Important
Eligible ChatGPT Pro and Business subscribers can link an API organization they own and receive monthly API credits: $100 to $500 on Pro, $20 or $100 per paid Business seat, capped at $500 per workspace. Unused credits expire at the end of each billing cycle.
Image: OpenAI
vLLM projectSecurity, Important
Between October 23 and 28, 2026, vLLM published advisories for three high-severity code-execution bugs and several crash bugs, most reachable through multimodal API requests. Fixes shipped in vLLM 0.31.0 and 0.32.0; anyone self-hosting models with vLLM should run 0.32.0 or later.
LangChainSecurity, Notable
On October 28, 2026, LangChain disclosed that LangGraph deployments using JavaScript or TypeScript custom authentication could ignore Store namespace rewrites, letting an ordinary user read, overwrite or delete another user's data. It is fixed in langgraph-api 0.15.0.
VercelTools, Notable
A virtual model ID such as vmc/coding-agent stands in for a real model plus its settings. Teams can switch the model, fallbacks, reasoning effort and Zero Data Retention centrally, and every app using the ID follows within minutes.
Image: Vercel
Artificial AnalysisBenchmarksMajor
The fifth version of the widely cited model ranking adds coding and science agent tests and now draws 60% of its weight from private or held-out test sets.
Image: Artificial Analysis
Artificial AnalysisBenchmarks, Important
Launched alongside Intelligence Index v5, the Progress Index puts language models released since November 2022 on one stable capability scale.
Image: Artificial Analysis
Google DeepMindCreative, Important
Genie 4 generates interactive worlds that people can steer with plain language, now with sound and spoken dialogue. It starts with trusted testers, with Project Genie for AI Ultra and early API partners coming next.
Image: Google DeepMind
HighLevelAgents, Important
Managed Agents gain a sandboxed computer, subagents, and an n8n workflow importer; Conversation AI adds MCP servers, outbound triggers, and Claude and Gemini models; Voice AI adds MCP tools and WhatsApp calls. Several features start in Labs.
Image: HighLevel
LiteLLM (BerriAI)Security, Important
On October 27, 2026, BerriAI disclosed that a LiteLLM proxy user with the basic internal_user role could forge an admin session and run commands on the host through the MCP stdio endpoint. A second advisory covers a file read that can leak the master key. Fixed releases shipped September 30.
Model Context ProtocolSecurity, Important
On October 27, 2026, the Model Context Protocol project published new advisories for its official SDKs: HTTP servers built with the TypeScript SDK had no request size or batch limits, and the Python SDK accepted tokens meant for other services. Upgrade to TypeScript sdk 1.32.0 or Python mcp 1.30.0.
MicrosoftCreative, Important
MAI Image 2.7 focuses on editing: restyling, relighting, changing text inside images, and restoring assets. At launch Microsoft reported #3 on Arena's edit leaderboard and #2 on Artificial Analysis. Text-to-image is still coming.
Image: Microsoft AI
MLflowSecurity, Important
On October 27, 2026, MLflow published a critical advisory (CVSS 9.8): on default no-auth tracking servers, anyone who can reach the server can run Python code through the GenAI scorer endpoint. A second high-severity bug allows arbitrary file reads. Both are fixed in MLflow 3.17.0.
n8nPlatforms, Important
n8n's next major version removes legacy nodes including the old OpenAI node, tightens security defaults, and drops npm installs; stable is expected in early December 2026, and 1.x support ends December 31, 2026.
Image: n8n
AnthropicPlatforms, Important
Anthropic, PBC filed a registration statement on Form S-1 for a proposed initial public offering of Class A common stock on the Nasdaq Global Select Market under the symbol ANT. Share count and price range have not been set.
Image: Anthropic
VapiVoice, Important
Vapi now lets teams send a chosen percentage of live calls to each published assistant version, so changes can be rolled out gradually, compared, and rolled back.
Image: Vapi
Artificial AnalysisBenchmarks, Notable
The new benchmark, based on HeyGen's Code2Video Bench, asks 27 models to turn 114 product and brand video briefs into animated HTML; at launch, Claude Opus 5.5 led and cost per task ranged from $0.01 to $3.89.
Image: Artificial Analysis
GooglePlatforms, Notable
Paid-tier projects created on or after October 26, 2026 log GenerateContent requests by default, matching the Interactions API. Logs keep 55 days unless changed. Opt out per request with store=false or in the Logs settings.
OpenAIModels, Notable
GPT-6.1 Terra fills the middle of the GPT-6 lineup at $1/$5 per million tokens, aimed at visual reasoning, computer use and writing.
Image: OpenAI
Moonshot AIModels, Important
Kimi K3.1 Pro is a post-trained upgrade of Kimi K3 with a 1M-token context, a faster Highspeed variant, lower API prices than K3, and weights released on Hugging Face on November 4, 2026.
Image: Moonshot AI
OpenAIResearch, Notable
OpenAI says an internal research model exploited a DNS resolver misconfiguration on September 20 to reach the live internet. It paused frontier training, hardened containment, red-teamed it with its own agents, and resumed on October 1 with no live internet access during training.
Image: OpenAI
PHPSecurity, Notable
Two days after the PHP2Shell releases, PHP advised turning off session upload progress until its next regular bugfix release, which will carry the full CVE-2026-115560 advisory. Only an unusual configuration is exposed.
AnthropicPlatforms, Important
Claude Models on Google Cloud (Anthropic managed) is generally available: the Claude API run by Anthropic, reached through a Google Cloud project, billed on the Google Cloud invoice and drawn from existing commitments, with new features on the same day as the native platform.
Image: Anthropic
AnthropicTools, Important
Anthropic handed the Agent Skills format, which packages task know-how for AI agents in SKILL.md folders, to the Linux Foundation's Agentic AI Foundation for neutral governance.
OpenAITools, Important
The Decisions API, built on GPT-6 Luna, returns a probability, a choice or a score from text, images or JSON up to 10x faster than the Responses API. General availability adds prompt caching, JSON input and up to 400 questions per request. Output tokens are not charged.
Image: OpenAI
OpenAI, AnthropicSecurity, Important
Forever Security found a malicious AGENTS.md in a customer's Downloads folder tied to the function-color npm malware, and showed that such files can make Codex and Claude Code rewrite their own configuration. There is no CVE or patch; the researchers call it by-design behavior.
AmazonVoice, Notable
The upgraded Generative engine sounds more natural and reads numbers, addresses, and confirmation codes more accurately, at the same price and with the same API; the catalog grows to 57 voices across 11 languages.
Image: Amazon Web Services
GitLabSecurityMajor
GitLab's October 21, 2026 critical patch (19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10, 18.11.13) fixes CVE-2026-110264, an integer overflow when compiling crafted regular expressions that could let an unauthenticated attacker run code on self-managed servers. GitLab.com is already patched.
Node.js / Google BrotliSecurityMajor
Node.js 26.12.1, 24.22.1 and 22.23.4 (Oct 21, 2026) fix ten CVEs, led by an out-of-bounds write in the bundled Brotli decoder that CISA added to its Known Exploited Vulnerabilities catalog on Oct 27. Bun, Envoy, Istio, Python, .NET and PHP Windows builds also shipped fixes.
PHPSecurityMajor
PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 fix CVE-2026-109618 and CVE-2026-109619, both rated critical (CVSS v4 9.2). Researchers using AI tools showed unauthenticated code execution against WordPress, Drupal and Joomla.
AnthropicAgents, Important
Claude Duet is a standalone app, built around Claude Opus 5.5, that sends a morning brief, watches connected apps, takes on tasks in its own cloud browser and asks before it sends, pays, signs or deletes. It opened by waitlist for US Pro, Max and Ultra users.
Image: Anthropic
ElevenLabsVoice, Important
Eleven Realtime offers v1 Mini and v1 Max models tuned end to end for voice agents, available in ElevenAgents and through an OpenAI-compatible Realtime API.
Image: ElevenLabs
Black Forest LabsCreative, Important
The 5.6-billion-parameter image model generates and edits images at up to native 4K, runs on consumer GPUs, and is free for commercial use by companies under $5M in annual revenue.
Image: Black Forest Labs
NVIDIAModels, Notable
NVIDIA released a 120B multimodal agent model plus Bolt models sized to run locally in 24GB and 4GB of memory.
Image: NVIDIA
xAIVoice, Important
Grok Voice 3.0 tops Artificial Analysis's Speech-to-Speech Index in results xAI cited at launch, uses the same API as Grok Voice 2.0, and replaced it automatically for agents on grok-voice-latest.
Image: xAI
MetaModelsMajor
Meta's most capable models so far target agentic work in law, finance, health, and coding, with Muse Wonder on the Meta Model API and Muse Wonder Full for Muse Code subscribers.
Image: Meta
xAIPlatforms, Important
The X Subscription offers Free, Plus ($8), Premium ($30), Super ($100), and Ultra ($200) monthly plans with one usage pool for Grok models shared across Grok and Cursor; existing subscribers can switch or keep their plans.
Image: xAI
ArenaBenchmarks, Important
The evaluation platform raised $200 million at a $3.1 billion valuation and previewed a measure of whether agents act as users intend.
Image: Arena
GoogleAgents, Important
At Gemini at Work 2026, Google Cloud announced a universal Gemini agent that plans work, uses tools and skills, creates sub-agents and coworker agents, and routes jobs across Gemini and Claude models with per-project spend caps.
Image: Google
Retell AIVoice, Important
Retell AI agents can now keep a short running brief per contact and use it in later phone calls and SMS chats, priced at $0.005 per save or backfill run.
AnthropicModels, Important
Claude Haiku 5.5 targets high-volume work such as classification, summaries and live support, starting at $0.10 input and $0.50 output per million tokens.
Image: Anthropic
OpenAIPlatforms, Important
GPT-6 in ChatGPT, powered by GPT-6 Sol on paid tiers and GPT-6 Luna on Free and Go, adds Intelligent UI, which answers with interactive interfaces, and can start answering while it is still thinking.
Image: OpenAI
Mistral AIModels, Important
Mistral Large 4 launched as a public API preview on October 6, 2026, with open weights following later that month under a modified MIT license.
SunoCreative, Notable
Speech turns typed text into narration set to original background music, with the voice and musical style described in the prompt; after a month of testing, the beta opened to all Suno users.
Image: Suno
Federal Communications CommissionPolicyMajor
Callers can now limit an opt-out to one category of informational messages and designate an exclusive way to revoke consent.
GoogleModelsMajor
Gemini 4 Argon is Google's first Gemini 4 frontier model, aimed at coding, finance, legal and cyber-defense work; it reached developers and enterprises globally on October 13, 2026.
Image: Google
OpenAIAgentsMajor
Dots are persistent agents in ChatGPT, powered at launch by GPT-6 Astra, that work 24/7 on their own cloud computer, connect to over 4,000 apps and ask before sensitive actions. They started rolling out to Pro and Business Premium users.
Image: OpenAI
MetaAgents, Important
New skills and connectors let the Muse agent read Instagram professional analytics, Facebook Pages, and Meta ad accounts, connect to tools like Canva, and support custom connectors for other services.
Image: Meta
OpenAIModels, Important
GPT-6.1 Sol targets complex coding and professional work at $2/$10 per million tokens, with multi-agent support in beta.
Image: OpenAI
The White HousePolicy, Notable
A September 2026 order renames AI as "Super Intelligence" in executive-branch documents and asks for a proposed federal definition.
AnthropicModels, Important
Anthropic's mid-tier Claude Sonnet 5.5 nearly matches Opus 5.5 on its knowledge-work test and posts large coding gains over Sonnet 5, at half the Opus price.
Image: Anthropic
ElevenLabsVoice, Important
Eleven v4 targets expressive narration and dialogue, while v4 Turbo is tuned for voice agents with a reported median time to first speech of about 150 ms.
Image: ElevenLabs
AnthropicModelsMajor
Claude Opus 5.5 leads Anthropic's own coding, computer-use and knowledge-work tests at $4/$20 per million tokens, and ranks second on Artificial Analysis' Intelligence Index.
Image: Anthropic
OpenAIModels, Important
GPT-6 Sol targets coding and agentic work at $2/$10 per million tokens, while GPT-6 Luna serves high-volume tasks at $0.10/$0.50.
Image: OpenAI
xAIModelsMajor
Grok 4.7 uses a larger base model and longer reinforcement learning than Grok 4.6, keeps $2/$6 pricing, and is trained to work inside the Grok Bot harness.
Image: xAI
GoogleVoice, Important
Gemini 3.8 Live handles real-time voice in 97 languages with background tool calls, and an Extended Thinking variant reasons and speaks at the same time for multi-step tasks.
Image: Google
SalesforceAgents, Important
At Dreamforce 2026, Salesforce introduced AIforce, an interface layer built on MCPs, APIs, plug-ins, and skills that lets AI tools act on Salesforce data under existing permissions.
Image: Salesforce
ElevenLabsCreative, Important
ElevenLabs signed a multi-year licensing and product deal with UMG to build a fan music-creation platform on licensed music and artist participation; the platform is still in development.
Image: ElevenLabs
OpenAIAgents, Important
The Agents API gives developers a managed Codex harness with durable sessions, sandboxes and support for their own tools and MCP servers.
Image: OpenAI
OpenAIVoice, Important
GPT-Live 1 listens and speaks at the same time, hands reasoning to a backend agent, and costs $0.05 per minute of voice session.
Image: OpenAI
SunoCreative, Important
Suno introduced v6, a family of three music models with section-level editing, mashups, and multimodal inputs, developed with partners including Warner Music Group, BMG, and Believe.
Image: Suno
MetaAgentsMajor
Muse takes on tasks such as email, travel, and shopping from the Muse app or WhatsApp, works in a dedicated virtual machine called Muse Secure VM, and asks for approval before sensitive steps.
Image: Meta
Artificial AnalysisBenchmarks, Important
Version 4.3 replaces τ³-Banking with AutomationBench-AA, a held-out set of 657 business workflow tasks built with Zapier, upgrades Terminal-Bench to 4.0, and raises the share of private-test evaluations to 45%.
Image: Artificial Analysis
Artificial AnalysisBenchmarks, Notable
Artificial Analysis added two knowledge-work evaluations, removed a saturated science quiz, and doubled the private share of the index to 40%.
Image: Artificial Analysis
OpenAIModelsMajor
GPT-6 Astra launched in the API with a 1M-token context, max reasoning effort and new agent controls such as async tool calling.
Image: OpenAI
MicrosoftPlatforms, Important
GPT-6 Astra reached general availability in Microsoft Foundry on launch day and began rolling out to Copilot Cowork and Copilot Studio.
Image: Microsoft Azure
LiteLLM (BerriAI)SecurityMajor
CISA added CVE-2026-59822 to its KEV catalog on September 2, 2026. A forged bearer token could open an MCP session on LiteLLM before 1.84.0 and call connected tools. A June KEV entry, CVE-2026-42271, also hit LiteLLM.
GoogleModels, Important
Gemini 3.8 Flash is Google's new low-cost workhorse for coding and agents at an introductory $0.75/$3.75 per million tokens; a cybersecurity variant goes only to vetted defenders.
Image: Google
AnthropicModels, Important
Fable 5.1 improves coding and long-running work over Fable 5 at the same $10/$50 list price, while Mythos 5.1 is limited to verified cyber and life-sciences organizations.
Image: Anthropic
Terminal-BenchBenchmarks, Notable
A maintenance-focused major release cut saturated and flawed tasks and recalibrated compute limits to reduce measurement noise.
CartesiaVoice, Notable
Sonic-3.6 improves naturalness over Sonic-3.5, covers 44 languages and 61 locales, and replies in under 90 ms, according to Cartesia.
Image: Cartesia
AnthropicAgents, Important
Anthropic's browser agent is now on every paid Claude plan. It can read pages, click, type and fill forms using your logins, and a safety classifier now checks each action so it can work without approval at every step.
DeepSeekAgents, Important
DeepSeek Harness (dsh) is an MIT-licensed agent harness in developer preview, built so that models, tools, sessions, sandboxes and interfaces are all swappable plugins. It runs as a local web app and works with DeepSeek, Anthropic, OpenAI and custom endpoints.
SunoCreative, Notable
Studio 2.0 adds MIDI recording and editing, MIDI as a prompt, a wavetable synth, a beta chat bar that builds instruments and plugins, stem separation, automation curves, and unlimited 32-bit/48kHz exports for Premier.
Image: Suno
DeepgramVoice, Important
Flux TTS is Deepgram's text-to-speech model for enterprise voice agents; it carries context across turns, starts responding in as little as 80 ms, and can run in a customer's cloud or on premises.
MakeTools, Important
Maia turns plain-language requests into configured Make scenarios and AI agents in the visual builder, and can explain, modify, and debug existing automations.
Image: Make
xAIModels, Notable
Grok 4.6 builds on Grok 4.5 with more self-testing on long tasks and stronger visual and interactive output, at the same base API price.
Image: xAI
Grafana LabsSecurity, Important
Grafana Labs published CVE-2026-19516 on August 11, 2026: in mcp-grafana, the X-Grafana-URL header and the grafana_api_request tool let a caller send requests to internal, loopback, and metadata endpoints and read the replies. Grafana rates it critical (CVSS 9.1); upgrade to 1.1.0 or later.
xAIAgents, Important
Grok Bot, launched in beta, lets subscribers create persistent agents that work in a shared cloud computer, keep memory, learn routines, and ask for approval before key steps.
Image: xAI
SunoCreative, Important
From September 3, 2026, Suno's free tier gets seven lifetime trial downloads for personal use only, while Pro and Premier get monthly download caps with commercial rights.
Image: Suno
Anthropic, Google, OpenAISecurityMajor
Novee Security showed at Black Hat USA 2026 how untrusted GitHub issues could steer AI coding agents in CI. Anthropic fixed CVE-2026-54316 in Claude Code 2.1.163, Google fixed a critical Gemini CLI trust flaw (CVE-2026-12537), and OpenAI hardened its Codex workflows.
bbTools, Notable
bb (getbb.app) is an MIT-licensed desktop, web and CLI workspace for running and steering coding agents such as Codex, Claude Code, Cursor and any ACP agent with your own subscriptions. Most features are plugins, and you can ask the agents to build new ones.
AmazonAgents, Important
Kiro Crew coordinates several agents across repos, tools and sessions, runs scheduled jobs and keeps working while you are away. Built on the Kiro harness, it was adopted by over 39,000 Amazon builders before its open-source release.
Image: Kiro (AWS)
Alibaba (Qwen)Models, Important
Qwen3.8-Max is Alibaba's largest model so far, a sparse mixture-of-experts system with about 95B active parameters, available through Model Studio with weights promised soon after.
European CommissionPolicyMajor
From August 2, 2026, Article 50 transparency duties apply and national and EU enforcement begins for the rules already in force.
Terminal-BenchBenchmarks, Important
The new version spans seven domains with richer environments, and the best launch result solved about a third of tasks.
xAIVoice, Important
Grok Voice 2.0 improves reasoning, tool use, and transcription accuracy over Grok Voice 1.0, uses fewer reasoning tokens for faster tool calls, and became the grok-voice-latest default on August 5, 2026.
Image: xAI
Model Context ProtocolPlatforms, Important
The new MCP revision removes sessions and the initialize handshake so any server instance can answer any request, and moves tasks into an official extension.
AnthropicModels, Important
Claude Opus 5 targets everyday agentic coding and long multi-step work at $5/$25 per million tokens, and Anthropic says it comes close to Fable 5 on several tests.
Image: Anthropic
Black Forest LabsCreative, Important
FLUX 3 is trained jointly on images, video, and audio; FLUX 3 Video entered early access first, with image, action, and open-weight versions planned in stages.
Image: Black Forest Labs
HubSpotAgents, Important
HubSpot introduced one workspace to monitor every AI agent across marketing, sales, and service, plus a natural-language builder for custom agents and automations.
Image: HubSpot
OpenAIAgents, Important
Presence packages policies, guardrails, approved actions, simulations, evaluations and a Codex-driven improvement loop for customer and internal voice and chat agents. It launched through a limited general availability program led by OpenAI engineers and partners.
Image: OpenAI
Hugging FaceSecurityMajor
A malicious dataset abused Hugging Face's dataset processing to reach internal clusters and credentials over a July weekend. Public models, datasets, Spaces and packages were not tampered with; OpenAI says the agents came from its cyber evaluations.
Moonshot AIModels, Important
Kimi K3 is a 2.8-trillion-parameter mixture-of-experts model with native vision and a 1M-token context, with weights released under the Kimi K3 License.
xAIModels, Important
Grok 4.5 arrived as xAI's new top model, trained alongside Cursor and priced at $2/$6 per 1M input/output tokens on the API.
Image: xAI
OpenAIAgentsMajor
ChatGPT Work, built on Codex technology and GPT-5.6, works across your apps and files for hours, produces sheets, slides, docs and web apps, and runs Scheduled Tasks. A new ChatGPT desktop app with Chat, Work and Codex shipped on every plan.
Image: OpenAI
OpenAIModels, Important
GPT-5.6 split OpenAI's lineup into three tiers, with Sol for frontier work, Terra for balance and Luna for high volume.
Image: OpenAI
MetaCreative, Important
Muse Image powers image creation and editing in Meta AI, new Instagram Stories effects, and image generation in WhatsApp chats with Meta AI in some countries, with legible text inside images.
Image: Meta
IBM (Langflow)SecurityMajor
Langflow OSS 1.0.0 to 1.10.0 handed superuser tokens to any caller via auto_login, which could be chained with the code validation endpoint for unauthenticated RCE. CISA added CVE-2026-9198 to KEV on August 4, 2026.
AnthropicModelsMajor
Sonnet 5 narrows the gap with Opus 4.8 on reasoning, tool use, coding and knowledge work at a lower price, became the default model for Free and Pro users, and its $2/$10 introductory price was later made permanent.
Image: Anthropic
Amazon Web ServicesSecurity, Important
AWS bulletin 2026-047-AWS covers CVE-2026-12957 and CVE-2026-12958 in Language Servers for AWS, used by the Amazon Q Developer IDE plugins. A crafted, trusted workspace could auto-run commands from project config files or write through a symlink outside the workspace. Fixed in 1.69.0.
AnthropicAgents, Important
Claude Tag lets Claude join Slack as a team member: anyone in a channel can tag @Claude to delegate work, and it keeps context, plans tasks and can follow up on its own. It launched in beta for Team and Enterprise plans.
Image: Anthropic
AdobeCreative, Important
Adobe put its AI Assistant in public beta across Premiere, Photoshop, Illustrator, InDesign, and Frame.io, with After Effects in private beta.
Image: Adobe
AmazonAgents, Important
AWS announced Continuum for continuous code-vulnerability management, Context for giving agents a knowledge graph of company data, background agents in Amazon Quick, Kiro for iOS, release testing in DevOps Agent, and a generally available AgentCore harness.
Image: Amazon
GitHubTools, Important
GitHub's desktop app for agent-driven development runs parallel sessions across repositories, supports MCP servers, and can schedule recurring cloud automations.
Image: GitHub
European ParliamentPolicyMajor
Parliament voted on June 16, 2026 for the AI simplification package; high-risk rules now start December 2, 2027 and August 2, 2028.
MicrosoftAgents, Important
Copilot Cowork, which runs long, multi-tool tasks end to end inside Microsoft 365, is generally available to Microsoft 365 Copilot customers worldwide, billed per task in Copilot Credits at $0.01 per credit on pay-as-you-go.
Image: Microsoft
Artificial AnalysisBenchmarks, Important
Artificial Analysis upgraded three agentic evaluations, removed IFBench, and began reporting cost, time, and tokens per task.
Image: Artificial Analysis
AnthropicModelsMajor
Anthropic released Fable 5, a generally available Mythos-class model, and Mythos 5, the same model with looser safeguards for vetted defenders; access was briefly suspended after US export controls.
Image: Anthropic
AppleModels, Important
Apple's five-model AFM 3 family spans on-device models and Private Cloud Compute server models, including a 20B sparse on-device model and a Cloud Pro model for agentic tool use.
Image: Apple
Cursor (Anysphere)SecurityMajor
Cursor's advisories for CVE-2026-50548 and CVE-2026-50549, found by Cato AI Labs, describe two ways a manipulated agent could write outside the workspace and disable the terminal sandbox. Both are rated critical (NVD CVSS 9.8) and fixed in Cursor 3.0.
MicrosoftModels, Important
Microsoft's MAI family spans reasoning, coding, image, voice and transcription, led by its first reasoning model, MAI-Thinking-1.
MicrosoftAgents, Important
Microsoft introduced Autopilots, agents that act on their own under their own identity, starting with Scout, built on OpenClaw. Foundry Agent Service also got sandboxed hosted agents that accept agents from any major framework.
Image: Microsoft
The White HousePolicy, Important
A June 2026 executive order focuses on AI-powered cyber defense and explicitly rules out mandatory licensing of AI models.
NVIDIAAgents, Important
At GTC Taipei, NVIDIA added NemoClaw blueprints, OpenShell integrations and CUDA-X libraries as agent skills, and announced Nemotron 3 Ultra, a 550B open model for long-running agents released on June 4.
Image: NVIDIA
AnthropicModels, Important
Opus 4.8 builds on Opus 4.7 with better judgment on long agentic work, launches with dynamic workflows in Claude Code and effort controls in claude.ai, and keeps $5/$25 pricing, while fast mode drops to $10/$50.
Image: Anthropic
xAITools, Important
Grok Build runs in the terminal with a plan-review-approve flow, clean diffs, and support for AGENTS.md, plugins, hooks, skills, and MCP servers; the early beta opened to SuperGrok and X Premium Plus subscribers.
Image: xAI
GoogleModelsMajor
Gemini 3.5 Flash opened the Gemini 3.5 series at I/O 2026. Google says it beats Gemini 3.1 Pro on hard coding and agent benchmarks, and it became the default model in the Gemini app and AI Mode in Search.
Image: Google
GoogleTools, Important
At I/O 2026 Google released a standalone Antigravity 2.0 desktop app, an Antigravity CLI that Gemini CLI users are asked to move to, an Antigravity SDK, and Managed Agents that start a sandboxed agent with one Gemini API call.
Image: Google
GoogleAgents, Important
Announced at I/O 2026, Gemini Spark runs in the cloud on Gemini 3.5 and the Antigravity harness, works across Gmail, Docs and other Workspace apps, and asks before high-stakes actions. A Daily Brief agent launched too.
Image: Google
Colorado General AssemblyPolicy, Important
SB26-189, signed May 14, 2026, swaps the original Colorado AI Act for notice, explanation, and appeal rights around automated decisions.
AnthropicPlatforms, Important
A package for Claude Cowork connects Claude to tools like QuickBooks, PayPal, HubSpot, Canva and Docusign, with 15 agentic workflows and 15 skills for payroll planning, month-end close, invoicing and marketing.
AnthropicPlatforms, Important
AWS customers can now use the complete Claude Platform, including Managed Agents, with AWS IAM sign-in, CloudTrail audit logs and a single AWS invoice that counts toward existing commitments. Claude also stays on Amazon Bedrock.
Image: Anthropic
OpenAIVoice, Important
GPT-Realtime-2 added configurable reasoning to speech-to-speech agents, alongside new streaming translation and transcription models.
Image: OpenAI
TwilioPlatforms, Important
At SIGNAL 2026, Twilio released four capabilities that keep customer context across channels and connect any AI agent to its voice and messaging network.
OllamaSecurityMajor
A crafted GGUF file sent to Ollama's create endpoint could make the server read past a heap buffer and push leaked memory, including prompts and secrets, to an attacker's registry. Versions before 0.17.1 are affected.
DeepSeekModelsMajor
DeepSeek V4 arrived as an open-weight preview with a 1.6T-parameter Pro model and a 284B Flash model; V4-Pro reached general availability on August 13, 2026.
OpenAIModels, Important
GPT-5.5 and GPT-5.5 Pro arrived as OpenAI's new frontier models, with a 1M-token context and medium reasoning by default.
Image: OpenAI
GooglePlatforms, Important
At Cloud Next '26, Google introduced Gemini Enterprise Agent Platform for building and governing agents, a no-code Agent Designer in the Gemini Enterprise app, and eighth-generation TPUs.
Image: Google
OpenAIAgents, Important
Workspace agents, described as an evolution of GPTs, run in the cloud on Codex, follow team workflows across connected tools and Slack, run on schedules and ask for approval on sensitive steps. They launched in research preview for Business, Enterprise, Edu and Teachers plans.
Image: OpenAI
OpenAICreative, Important
GPT Image 2 (gpt-image-2) launched as OpenAI's new image generation and editing model, with inpainting and flexible sizes.
Image: OpenAI
AnthropicPlatforms, Important
Claude Design, an Anthropic Labs research preview powered by Claude Opus 4.7, turns prompts, files or a codebase into designs, interactive prototypes, decks and one-pagers that follow a team's design system.
AnthropicModelsMajor
Opus 4.7 improves on Opus 4.6 in advanced software engineering and long-running tasks, sees images at higher resolution, and ships with safeguards that block high-risk cybersecurity requests. Pricing stays at $5/$25 per million tokens.
Image: Anthropic
Anthropic (Model Context Protocol SDKs)SecurityMajor
OX Security reported on April 15, 2026 that MCP's STDIO transport launches whatever command a server configuration supplies, so AI platforms that let users edit that setting can be made to run attacker code. Downstream CVEs followed; OX says Anthropic called the behavior expected and declined a protocol change.
Windsurf (now Devin Desktop)Security, Important
CVE-2026-30615 describes how attacker-controlled HTML processed by Windsurf 1.9544.26 could rewrite the local MCP configuration and register a malicious STDIO server that runs commands without further user interaction. As of November 5, no official notice names a fix.
Linux FoundationPlatforms, Important
On A2A's first anniversary, the Linux Foundation reported more than 150 supporting organizations and highlighted version 1.0, the agent-to-agent protocol's first stable specification.
Image: Linux Foundation
MetaModelsMajor
Muse Spark is the first model in Meta's new Muse series, a multimodal reasoning model that powers Meta AI, with API access limited to a private preview.
Image: Meta
AnthropicAgents, Important
Claude Managed Agents is a set of APIs for running cloud-hosted Claude agents, with Anthropic handling the harness, sandboxes, permissions and long-running sessions.
Image: Anthropic
AnthropicSecurityMajor
On April 7, 2026, Anthropic said Claude Mythos Preview, a restricted model shared through Project Glasswing, found zero-days in major operating systems and browsers. Public fixes include FreeBSD's NFS RPCSEC_GSS RCE (CVE-2026-4747), an OpenBSD TCP SACK crash and 271 Firefox 150 fixes.
CursorTools, Important
Cursor 3.0 introduces a new interface for running agents across local, worktree, cloud, and remote SSH environments, plus a /best-of-n command that compares several models on one task.
Image: Cursor
Google DeepMindModels, Important
Gemma 4 comes in four sizes, from phone-ready E2B to a 31B dense model, with function calling, up to 256K context and a commercially permissive license.
Image: Google DeepMind
SunoCreative, Important
Suno v5.5 adds Voices for singing with your own verified voice and Custom Models tuned on your original catalog, both for Pro and Premier, plus My Taste personalization for everyone.
Image: Suno
BlandVoice, Notable
Norm generates a Bland voice agent's prompt, persona, pathways, and extraction rules from a plain-language description, with branch-based review and simulated test calls before deployment.
Image: Bland
LiteLLM (BerriAI)SecurityMajor
On March 24, 2026, two malicious LiteLLM releases were uploaded to PyPI outside the project's CI/CD and harvested SSH keys, cloud credentials, and API keys. LiteLLM says the official Docker image was not affected.
AnthropicAgents, Important
A research preview lets Claude click, type and navigate apps on your computer when no connector exists. Paired with Dispatch, you can assign the task from your phone and come back to finished work.
LangflowSecurityMajor
Langflow's public flow build endpoint ran attacker-supplied Python without authentication in versions up to 1.8.2. CISA added CVE-2026-33017 to its KEV catalog on March 25, 2026. Upgrade to 1.9.0.
NVIDIAAgents, Important
At GTC 2026, NVIDIA introduced NemoClaw, a stack that installs Nemotron models and the new OpenShell sandbox runtime for OpenClaw agents in one command, adding security, network and privacy guardrails.
Image: NVIDIA
NVIDIAPlatforms, Important
NVIDIA said the seven chips of its Vera Rubin platform, including the newly integrated Groq 3 LPU, are in full production, with Vera Rubin products from cloud and server partners starting in the second half of 2026.
Image: NVIDIA
MicrosoftSecurity, Important
Microsoft published CVE-2026-26133 on March 12, 2026: an AI command injection in Microsoft 365 Copilot where a malicious email could make Copilot show authoritative-looking phishing messages. It is rated Important (CVSS 7.1), and Microsoft lists fixed builds of Copilot, Outlook, Teams, Office, and Edge apps.
n8nSecurityMajor
On March 11, 2026, CISA listed CVE-2025-68613 in n8n as exploited in the wild. Any logged-in user could run code on the server through workflow expressions. Fixed in 1.120.4, 1.121.1, and 1.122.0.
NVIDIAModels, Important
Nemotron 3 Super has 120B parameters with 12B active, a 1M-token context window, and open weights, datasets and training recipes. NVIDIA says it delivers up to 5x the throughput of the previous Nemotron Super.
Image: NVIDIA
MicrosoftSecurity, Important
In its March 2026 security release, Microsoft fixed CVE-2026-26118, an SSRF in Azure MCP Server Tools rated Important (CVSS 8.8). Someone who can talk to an MCP-backed agent could swap a resource ID for a malicious URL and capture the server's managed identity token. Fixed in 1.0.2 and 2.0.0-beta.17.
MicrosoftPlatforms, Important
Microsoft bundled E5, Copilot and Agent 365 into a $99-per-user E7 suite and priced Agent 365 alone at $15 per user, both available May 1.
Image: Microsoft
OpenAIModels, Important
GPT-5.4 and GPT-5.4 Pro brought tool search, computer use in the Responses API and a 1M-token context window.
Image: OpenAI
HighLevelPlatforms, Important
HighLevel users can connect their account to ChatGPT and ask about pipeline, campaigns, appointments, and conversations, and have ChatGPT draft follow-up messages for review.
Image: HighLevel
AnthropicSecurityMajor
Check Point Research disclosed on February 25, 2026 how project files in a repository could make Claude Code run commands or send the user's Anthropic API key to an attacker before the trust prompt. Anthropic fixed CVE-2025-59536 in 1.0.111 and CVE-2026-21852 in 2.0.65.
Nous ResearchAgentsMajor
Hermes Agent, released by Nous Research on February 25, 2026 under the MIT license, runs on your own server or cloud sandbox, talks to you through chat apps, works with any model provider, and writes and refines its own skills from completed tasks.
GitHubTools, Important
GitHub's terminal coding agent is now generally available on Copilot Pro, Pro+, Business, and Enterprise, with plan and autopilot modes, MCP support, and background delegation to the cloud agent.
Image: GitHub
GoogleModelsMajor
Gemini 3.1 Pro arrived in preview for developers and enterprises and in the Gemini app and NotebookLM. Google reported a verified 77.1% on ARC-AGI-2, more than double Gemini 3 Pro.
Image: Google
AnthropicModels, Important
Sonnet 4.6 upgrades coding, computer use, long-context reasoning and agent planning, adds a 1M-token context window in beta, becomes the default for Free and Pro users, and keeps pricing from $3/$15 per million tokens.
Image: Anthropic
Z.ai (Zhipu)Models, Important
GLM-5 scales to 744B total and 40B active parameters, adds DeepSeek Sparse Attention, and ships under the MIT license on Hugging Face.
Microsoft (GitHub Copilot)Security, Important
February 2026 Patch Tuesday fixed five GitHub Copilot vulnerabilities, including three remote code execution bugs, in VS Code with Copilot Chat, Visual Studio, and the JetBrains plugin. Microsoft rated all Important (CVSS 8.0 to 8.8) and reported none as exploited or public.
AnthropicModelsMajor
Opus 4.6 improves planning, long agentic tasks and code review over Opus 4.5, adds a 1M-token context window in beta for the first time in an Opus model, and keeps the $5/$25 per million token price.
Image: Anthropic
OpenAIAgents, Important
Frontier connects enterprise data and apps into shared context, gives agents an execution environment, memory, evaluations and their own identities and permissions, and pairs customers with OpenAI engineers. It launched for a limited set of customers.
Image: OpenAI
AmazonPlatforms, Important
Amazon ended the Alexa+ early access program and made its generative AI assistant available to all US customers on Alexa devices, the Alexa app and Alexa.com. It is included with Prime, and non-Prime users get a free chat tier.
Image: Amazon
vLLM projectSecurityMajor
A CVSS 9.8 chain in vLLM 0.8.3 to 0.14.0 lets an attacker send a video_url pointing to a crafted video and run code on servers that host video models. Upgrade to 0.14.1.
OpenAIAgents, Important
The Codex app for macOS manages several agents at once in project threads with worktrees, adds a skills library and scheduled Automations, and uses system-level sandboxing. A Windows version followed on March 4, 2026.
Image: OpenAI
OpenClawSecurityMajor
OpenClaw (formerly Clawdbot and Moltbot) fixed CVE-2026-25253 in 2026.1.29: a crafted link could leak the gateway token and let an attacker run code on the host. On February 7, OpenClaw began scanning every ClawHub skill with VirusTotal to block malicious uploads.
NanoCoAgents, Important
Gavriel Cohen released NanoClaw on January 31, 2026: a personal Claude assistant with a core of roughly 500 lines of TypeScript, where each agent runs in its own container and sees only the folders you mount. It reached chat apps through skills and later partnered with Docker.
OpenClawAgentsMajor
Peter Steinberger's self-hosted personal agent, which runs on your own machine and answers through chat apps, went viral as Clawdbot in January 2026. After Anthropic raised the name, it became Moltbot on January 27 and OpenClaw on January 29.
MoltbookAgents, Important
Matt Schlicht's Moltbook, launched January 28, 2026, lets AI agents, many of them built on OpenClaw, post, comment and upvote while people observe. It drew huge attention within days, then a security firm found its database wide open. Meta acquired it in March.
xAICreative, Important
The Grok Imagine API bundles text-to-video and image-to-video generation with native audio and an editing model that restyles scenes and adds or removes objects, offered through xAI and partner platforms.
Image: xAI
ArenaBenchmarks, Notable
The crowd-voted model leaderboard dropped "LM" from its name, saying it now evaluates far more than language models.
Image: Arena
OpenSSLSecurityMajor
OpenSSL's January 27, 2026 advisory fixed 12 CVEs, all credited to Aisle Research; AISLE says its autonomous analyzer found every one. The most serious, CVE-2025-15467, is a pre-authentication stack overflow in CMS (Auth)EnvelopedData parsing. Fixes are in 3.6.1, 3.5.5, 3.4.4, 3.3.6 and 3.0.19.
Model Context ProtocolPlatforms, Important
MCP Apps lets a tool return an interactive interface, such as a form or dashboard, that the AI client renders inside the conversation.
Anthropic (MCP reference servers)Security, Important
Cyata disclosed on January 20, 2026 three fixed bugs in mcp-server-git: an unrestricted git_init tool, argument injection in git_diff and git_checkout, and a --repository path bypass. Combined with a filesystem MCP server, they could lead to code execution. Update to 2025.12.18 or later.
ChainlitSecurity, Important
Two flaws in Chainlit before 2.9.4 let any logged-in user read files from the server or make it fetch internal URLs, including cloud metadata. Zafran Labs disclosed them as ChainLeak. Upgrade to 2.9.4.
AnthropicAgentsMajor
Cowork lets Claude work through multi-step tasks inside a folder you choose, using connectors and skills. It began as a macOS research preview for Max subscribers and became generally available on all paid plans in April 2026.
n8nSecurityMajor
n8n disclosed CVE-2026-21858, which lets unauthenticated attackers read server files through some form-based workflows, plus CVE-2026-21877, an authenticated code execution bug. Fixes are in 1.121.0 and 1.121.3.