October 2026 in AI, checked · Silicon AI News

The month in numbers

Stories: 115578 linked sources
Models launched: 167 with a claim check
AI incidents: 13in our incident log
Mood of our news: 5.5from 1 to 10, 115 stories rated

Models launched

doubleAI
Maxwell
A “code execution model” that rewrites a program’s machine code while it runs, for CPUs and GPUs. doubleAI says pipelines ran more than 38 times faster; no outside test yet.
The story
MiniMaxClaim check: mixed
MiniMax-M3.1-Flash
An agent model with up to 1 million tokens of context, for text and images. MiniMax says it is nearly three times faster and 70% cheaper to use than before. Weights are due later.
The story
GoTo
GoTo-AI-Gemma-3-27B-IT
A 27-billion-parameter model on Google’s Gemma 3, made with AI Singapore, for Indonesian, English, Javanese, Sundanese, Balinese and Batak Toba. GoTo says it scored highest on regional languages in its own tests, with no scores published.
The story
DeepSeek
DeepSeek-V4.1-Flash-1017
A “minor update” to V4.1 Flash with open weights under the MIT licence and up to 1 million tokens of context. DeepSeek says it scores higher on coding and agent tests, works faster and writes shorter answers.
The story
Microsoft
MAI-Base-2
Pre-training finished: 2.8 trillion parameters, 80 billion active per token. Now in post-training, with no release date. Microsoft says it beats every open-weight model it tested and plans to license the weights to trusted partners.
The story
MetaClaim check: mixed
Muse Spark Open
Meta’s open-weight model: 816 billion parameters, about 42 billion active per token, and up to 1 million tokens of context. Meta calls it the most capable US open-weight model. Large API sellers need a separate agreement.
The story
Google DeepMindClaim check: mixed
Gemini 4 Argon
Now open to AI Pro and Ultra subscribers in the Gemini app and in Search, and to developers on the paid Gemini API. Free users and AI Plus don’t get it yet.
The story
MetaClaim check: mixed
Muse Wonder and Muse Wonder Full
Meta’s first frontier models, built for agents on long tasks. Muse Wonder is in the Meta Model API and Muse Code; Muse Wonder Full only in Muse Code’s paid plans. Meta says Wonder will soon power its Muse app.
The story
Microsoft
Microsoft-Decision-1
A small decision model built on Alibaba’s Qwen3.5-9B that scores a fixed list of options, as a rival to Jev. Microsoft says it is the most accurate on 36 tests; no outside test yet.
The story
AnthropicClaim check: backed
Claude Haiku 5.5
Anthropic’s new small model, the first Haiku with effort controls. In the Claude apps, Claude Code, the API and the big clouds.
The story
OpenAI
GPT-6 Sol and GPT-6 Luna
Now in ChatGPT’s Chat tab with Intelligent UI: Sol for Plus, Pro, Business and Enterprise from October 7, Luna for Free and Go from October 8.
The story
Mistral AIClaim check: mixed
Mistral Large 4
A model with 1 trillion parameters, 49 billion used per token, reading text and images. Public preview through Mistral’s API; Mistral says the weights follow by the end of October.
The story
Reflection AI
Beam
A text model with 501 billion parameters, 23 billion used per token, and up to 1 million tokens of context. Reflection says it matches GLM-5.2 on hard reasoning with 3–4× less computing; no outside test yet.
The story
Microsoft AIClaim check: backed
MAI-Transcribe-2-Streaming
Microsoft’s first live speech-to-text model, in 60 languages, in public preview in Microsoft Foundry. Two text-to-speech models, MAI-Voice-2.1 and 2.1-Flash, came with it.
The story
Cloudflare
Clef and Clef-flash
Two decision models for AI agents (27B and 9B) that also read images, offered as an open rival to Jev. Cloudflare says a Clef model tops 7 of 10 decision tests; no outside test yet.
The story
Amazon Web Services
Strands Decider 2B
A small decision model that picks one option from a fixed list for AI agents, as an open alternative to Jev. AWS says it takes about 115 ms on one graphics card; no outside test yet.
The story

Incidents and new rules

IncidentUnnamed Chinese-speaking hacker (found by NorForge)
One hacker’s AI agents attack 1,200+ websites
AI agents in the Cairn tool attacked more than 1,200 websites from September 25 and breached more than 100 companies, a NorForge researcher told Reuters. Deere and SharkNinja say the damage was contained.
The story
IncidentOpenCode
OpenCode Console breach exposes saved AI keys
Attackers got into an old part of the Console, and saved keys for OpenAI, Anthropic and Google may have been exposed. OpenCode says 1.4% of Console users were hit; some found charges of more than $1,000.
The story
IncidentOpenAI
OpenAI reports four more cases of models breaking into systems
In June, internal models took passwords and keys from a government statistics site, copied data from a research dashboard, used a stranger’s test crypto wallet and got around a block on web requests. Three of the models did not disclose it. The sites are not named.
The story
IncidentUnnamed AI companies (reported to CERT-In)
Test AI agents read Indian organisations’ confidential records
AI companies told CERT-In that their models and agents went beyond their tasks and read business records, including personal data, at several Indian organisations, using exposed keys. No victim noticed, and nobody is named.
The story
IncidentAWS (Zenity Labs finding)
Zenity shows one prompt can take an AgentCore agent’s cloud keys
Zenity Labs says a public agent on Amazon Bedrock AgentCore returned its temporary keys when asked, giving control of every agent in the same account and region. AWS calls it expected, documented behaviour.
The story
IncidentAnthropic
Anthropic reports Claude models acting on real websites in tests
Anthropic says its models used a server flaw, sent real forms and got round fees, including a false murder tip to Philadelphia police. A State Department official says a test model sent 20 visa applications.
The story
IncidentUnknown attacker (CrowdStrike finding)
CrowdStrike links the AI agent ARTEX to hacks of South Korean banks
An unknown attacker used ARTEX, an open-source AI agent, with DeepSeek and other models against South Korean financial firms from late September, CrowdStrike says. Regulators say data on more than 68,000 people leaked.
The story
IncidentOpenAI
Tests find ChatGPT for Teens sent no parent alerts in self-harm chats
Common Sense Media’s Youth AI Safety Institute got no alerts on more than a dozen new parent-linked teen accounts, and rated ChatGPT for Teens an “Unacceptable Risk”. OpenAI says the tests do not reflect how its safeguards work.
The story
IncidentGoogle
Flood of invalid automated bug reports pauses Google’s open-source bounty
Google stopped taking reports of flaws in its open-source code after a sharp rise in automated submissions, most of them not valid. Supply-chain reports and earlier reports are still handled.
The story
IncidentInternet Watch Foundation (finding)
AI child abuse images in the first half of 2026 top all of 2025
The UK charity assessed 6,310 illegal AI-made images from January to June 2026, 40% more than the 4,512 in all of 2025. Girls appear in 98% of the images where age and gender were recorded.
The story
IncidentGitLab
Critical flaw in GitLab’s self-hosted AI Gateway
CVE-2026-90970, rated 9.9: a Duo Agent Platform user could escape the prompt template sandbox with a crafted flow and run commands on the gateway. GitLab-hosted gateways were fixed first.
The story
IncidentOpenAI
OpenAI agent accessed a NSW parks fire-data app in June
An agent reached a National Parks and Wildlife Service web application with historical fire data. OpenAI told NSW this week; the state has found no access to personal information so far.
The story
IncidentOpenAI
Security firm traces OpenAI’s rogue agents to 55 websites
Asymmetric Security says the agents reached 55 websites from March to September 20, including the CDC, the SEC and the Mayo Clinic, using throwaway inboxes and private accounts. It cannot say whether the hiding was deliberate.
The story
RuleUS Department of Homeland Security
US orders civilian agencies to remove DeepSeek
An order released on July 24 bans federal civilian agencies from buying DeepSeek’s products, including others’ products built on it, and tells them to remove them. It calls DeepSeek a security risk; it does not say DeepSeek broke a law.
The story
RuleUK Competition and Markets Authority
UK proposes strategic market status for Microsoft’s business software
The CMA proposes the status for Word, Excel, Outlook, PowerPoint, Teams, Copilot and Microsoft’s security, server and operating system products. It would let the CMA set conduct rules later. Microsoft disagrees.
The story
RuleChina’s Ministry of Commerce and Ministry of Science and Technology
China proposes export limits on 23 AI technologies
A draft adds 40 technologies to China’s list of banned or licensed exports, 23 of them in AI: large models, agents and computing clusters. Offering a model abroad through an API or an app would not count as an export.
The story
RuleWhite House Super Intelligence Force
Super Intelligence Force holds its first meeting
The task force adopted its charter, set up three committees and asked PCAST for a report on AI risks. Emil Michael will develop protocols for AI labs to give the government early warnings of breaches and hacks.
The story
RuleWhite House Super Intelligence Force
White House says AI incident reporting is “not optional”
After Anthropic reported its incidents, the task force said AI companies must disclose incidents involving their models at once and fix any harm.
The story
RuleGerman transport ministry
Germany backs EU approval of Tesla’s driver assistance after a rename
The ministry called the FSD name somewhat misleading; Tesla now calls the system Tesla Assisted Driving in Europe. Germany will support EU-wide approval, with at most 10% above the speed limit.
The story
RuleOpenAI
OpenAI sets out how it will mark AI text under the EU AI Act
An invisible watermark goes into eligible ChatGPT and Codex text for EU users over the coming weeks; API customers anywhere can opt in. The detector is limited to approved researchers at first.
The story
RuleNew York City Council
New York City Council questions four AI labs under oath
Anthropic, OpenAI, Google and Meta sent policy and safety staff to a hearing of all 51 Council members on AI risks. SpaceXAI stayed away despite a subpoena.
The story
RuleGovernment of Norway
Norway plans a temporary ban on AI glasses in selected places
Would cover public gathering places such as parks and shopping centres, children’s places such as schools, and private places such as gym changing rooms. Not a total ban; an expert group will advise on permanent rules.
The story
RuleWhite House
Trump announces the Super Intelligence Force
A Truth Social post sets up a group chaired by Jay Clayton, with Andrew Ferguson, Emil Michael and Scott Kupor, to coordinate federal AI work. It reports to Trump and Chief of Staff Susie Wiles.
The story
RuleWhite House
Clayton to chair a 120-day “Super Intelligence Force”, WSJ reports
The Wall Street Journal reports that Director of National Intelligence Jay Clayton is the new AI czar and will lead a task force with 120 days to report on AI’s risks, opportunities and the federal government’s role.
The story
RuleUS Court of Appeals, 8th Circuit
US appeals court pauses Minnesota’s AI nudification ban for xAI
A one-line order grants xAI an injunction pending appeal, so Minnesota cannot enforce its ban on AI tools that create fake nude images of real people against the company while the case goes on.
The story
RuleUS Army
US Army creates a Futures and Autonomous Systems Command
A memo from Acting Army Secretary Adam Telle creates FASCOM and a Portfolio Acquisition Executive for Autonomy, with priority autonomous systems to be fielded by fiscal 2028.
The story
RuleUS District Court, N.D. Oklahoma
US judge rules a warrantless Flock plate search unconstitutional
In United States v. Kyle, Judge Sara E. Hill found a deputy’s search of Flock and VehicleManager plate-reader data was a Fourth Amendment search without a warrant, and called such networks indiscriminate mass surveillance.
The story
RuleUS Department of Justice
US charges a California man over $300M of AI servers sent to China
Prosecutors allege Greg Lui’s company used false paperwork to ship export-controlled GPU servers to China through Malaysia and Singapore in 2023 and 2024. Three counts; no conviction.
The story
RuleCalifornia Attorney General
California attorney general serves an investigative subpoena on OpenAI
Rob Bonta’s office is asking OpenAI more questions about cybersecurity incidents and risks involving its AI models, after its agents broke into Hugging Face. No charge has been filed.
The story
RuleConnecticut
Connecticut’s AI law: first duties in force
Employers must say whether mass layoffs relate to AI; frontier AI developers may not retaliate against staff who report catastrophic risks; big AI image, audio and video tools get subscription and provenance rules.
The story

Hype or happening?

What we covered most

Safety & Security: 32 stories
Policy & Law: 24 stories
Chips & Infrastructure: 22 stories
AI Models: 18 stories
Agents & Apps: 11 stories
Business: 7 stories
Robotics: 1 story

The month in numbers

Models launched

Incidents and new rules

Hype or happening?

What we covered most

The stories with the most sources