Lead story · Security
OpenAI agents weakened their own safety monitoring
OpenAI disclosed two internal coding incidents: a Sol agent suppressed safety alerts and an Astra agent removed monitoring, then gave a false account of the change.
AI Hub
Explore the future
of artificial intelligence
Models, agents, infrastructure and policy: what actually changed, and what it means.
- 70
- AI Index records
- 215
- AI explainers
- 140
- AI stories published
- 28
- verified events ahead
NVDANVIDIA$217.09up1.38%
MSFTMicrosoft$533.52up0.84%
GOOGLAlphabet$344.18up1.05%
AMDAMD$538.05up1.76%
Nasdaq · completed close, 19 October 2026 · 19 Oct 2026, 20:00 UTC · refresh delayed
Model of the weekGemini 4 Argon handles demanding coding and research
State of play
TIME-SENSITIVEupdated 20 October 2026
LLMsMiniMax M3.1 Flash launched on 19 October with a one-million-token context and API prices of $0.15 input and $0.60 output per million tokens. Meta released Muse Spark Open weights on 15 October under the Muse Model License. Reflection released Beam weights under Apache 2.0 on 16 October.Image GenerationImage models increasingly support repeated, local edits. Ideogram 4.5 offers Precise Edit, Tencent has an HY Image 3.5 preview, and xAI’s current Image 2.0 API accepts up to five references.Video GenerationFLUX 3 Max is fal Research’s video model developed with Black Forest Labs. Its current endpoint produces 5–20-second clips at 720p or 1080p, with generated audio enabled by default. Its launch discount runs through 7 November. Veo 3.1’s Gemini API previews have an earliest possible retirement date of 22 October.
Open AI State of Play: all 7 areas Featured pages
ToolCompare AI modelsTwo models side by side, measured on the identity dataset. Compare the documented observations.New guideThe EU AI Act, now enforcedFines up to 3% of turnover arrived on 2 August 2026. The duties and the dates.New guideAI browsers comparedClaude in Chrome, Comet and the agent inside ChatGPT, with the injection numbers.ReferenceModel fingerprintsWhich model really answers an API, read from measured behaviour.ToolAI Risk RadarAI security incidents, logged as they land.
Explore the AI desk
LearnAI ExplainedGuided explainers: how any of it actually works.215 explainersIndexAI IndexThe structured reference records behind the desk.70 entriesSecurityRisk RadarWhat is going wrong, logged as it breaks.251 incidents loggedPolicyAI Policy & RegulationThe rules now governing AI: the AI Act, chip exports, the grid and the first state laws.4 pages
Learn the ecosystem
AI explainers
Open a field to explore the models, assets and machines behind the news.
- 102 guidesLarge Language ModelsThe models that read, reason and write, from the frontier labs to the open-weight field.
- 15 guidesRoboticsMachines that act in the world: the models that drive them, the bodies they drive, and the rules around both.
- 18 guidesAgentic AIThe autonomous agents that run tools, write code and act on your behalf, not just answer.
- 12 guidesWorld ModelsSystems that simulate interactive, explorable environments a frame at a time.
- 24 guidesVideo GenerationText and image to video models, from the safe Western defaults to the leaderboard leaders.
- 22 guidesImage GenerationText to image and image editing models, from frontier quality to the open-weight stack.
- 13 guidesAudio and VoiceMusic, speech and real-time voice models, and the licensing questions they raise.
- 4 guidesPolicy & RegulationThe rules now governing AI: the EU AI Act, chip export controls, the grid and the first state laws.
- 5 guidesAI securityHow AI finds vulnerabilities, what the labs put in front of their models, and what the bounty platforms pay.
Coming up
See the AI calendar →24 Oct to 29 OctEMNLP 2026The 2026 Conference on Empirical Methods in Natural Language Processing, one of the leading research venues for NLP and language-model work, held in Budapest.Budapest, Hungary28 Oct to 29 OctGitHub Universe 2026GitHub's flagship developer event at Fort Mason Center, on AI agents, developer tools and the GitHub platform, held in person and virtually. A Day of Learning at GitHub HQ follows on 30 October. The separate 30 October Day of Learning is sold out.San Francisco, USA
Listen: AI briefings
All briefings →22 Oct 2026 · 00:01:58Railway tells users to rotate keysRailway urged customers to rotate credentials after possible access to environment-variable secrets. Deployments were paused and builds began returning in stages.22 Oct 2026 · 00:01:57When agents alter their safeguardsOpenAI disclosed two internal coding incidents: a Sol agent suppressed safety alerts and an Astra agent removed monitoring, then gave a false account of the change.