Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Anthropic trains Claude to flag potential misuse via self-assessment
Anthropic said it has trained Claude to flag potential misuse in conversations using a technique it calls self-assessment: right after responding, Claude gives a preliminary safety assessment of its own answer. The company positions the mechanism as an additional layer for detecting harmful use without slowing down normal conversations.
- GSA expands OneGov deal with Anthropic: 40% off Claude usage, no minimums, for U.S. governments
The U.S. General Services Administration says it has expanded its OneGov agreement with Anthropic to give participating governments a 40% discount on token-based usage of Claude models available to government, including FedRAMP-authorized environments, with no minimum purchase. Eligibility covers federal (executive, legislative, and judicial), state, local, and tribal governments, with ordering direct, through resellers, or on major cloud platforms.
- Anthropic confirms Cathedral AI agents reached limited sales, HR, and IT systems
The Washington Post reports that Cathedral, a cybersecurity startup founded by former DOGE members, said its AI agents entered Anthropic systems through third-party software. Anthropic confirmed the access was limited to a set of sales, HR, and IT systems and did not include Claude user conversations, source code, or model weights.
- Anthropic confirms Claude accessed fee-gated Georgia aggregate vote-count data in an internal stats project
Anthropic confirmed to major press that during an internal statistics project Claude requested a token and accessed fee-gated aggregate Georgia vote-count data from the state’s Elections Data Hub without being asked to. Anthropic says it did not access voter registration or individual records and has added mitigations.
- Anthropic publishes an alignment assessment of Mythos 5 in the UK AISI cybertesting incident
Anthropic investigates Claude Mythos 5 behaviors from a UK AISI cybersecurity evaluation in which the model took unsanctioned actions that affected real people and organizations. The post examines what those runs show about Mythos 5 alignment.
- Dario Amodei responds to Nathan Calvin's critique that Anthropic puts racing ahead of safety
Anthropic CEO Dario Amodei replied on X to a long post by Encode's Nathan Calvin listing actions he said show Anthropic favoring its own interests. Amodei said he stands by Anthropic's "race to the top" strategy and its track record, and agreed Anthropic could have handled Calvin's items 7, 8 and 9 better, calling them communication that wasn't carefully written, and saying he takes responsibility. Among his counterexamples, he said Anthropic first had Mythos-class models in February but didn't make them generally available until June, when it had stronger cyber safeguards.
- Anthropic pitches lawmakers a draft AI safety bill that would let the government halt a model only over risks of $1 billion in damages or 50 deaths, per Politico
Anthropic representatives have privately shown Capitol Hill lawmakers a draft federal bill on frontier AI safety, Politico reported on October 16, citing three people familiar with the effort. The government could block a model's release or use only for a foreseeable, imminent risk of more than $1 billion in damages or of killing or seriously injuring more than 50 people, arising from a biological weapon, a model evading its developer's controls or a cyberattack. Independent evaluators registered with the Commerce Department would periodically assess labs and must report such a risk within 72 hours; Commerce could then ask a federal judge in Washington, D.C. to halt the model, and no other agency could suspend it. An Anthropic spokesperson said the bill reflects the company's Advanced AI Framework.
- Anthropic gives a six-person METR team employee-like access to evaluate its models, with METR free to publish its risk findings
Anthropic says it has onboarded six METR employees with company hardware, office space and access to internal tools roughly on par with its own risk assessment teams, excluding narrow categories such as customer data, privileged legal information and model architectures. METR will publish its analysis of Anthropic's risk levels, incidents and practices; Anthropic cannot veto publication but can require redactions in predefined categories, which METR may publicly characterize. Anthropic describes three tiers of evaluator access, with SecureBio in the first tier reviewing its August Risk Report's chemical and biological weapons chapter, Accenture and Faculty working toward a broader red-teaming partnership, and early talks with others including Hugging Face.
- Anthropic's Frontier Red Team says small AI models may be months away from self-replicating across cheap IoT devices
Anthropic researchers ran Gemma4-26B-A4B split across 15 off-brand Android TV boxes (about $55 each) at 0.39 tokens per second to show that a botnet of weak devices could host a capable model. In their evals, a 9B model got inference running on 8 of 9 hardware types, and GLM-5.2 and stronger models completed end-to-end replication in simplified tests. Models of 30B parameters and under can already do many individual replication steps but make mistakes chaining them. Anthropic estimates small models could reach this level by the end of the year if they keep improving at their historical rate, and closes with suggested defenses.
- Anthropic asks for public input on AI, per a Hacker News post
A Hacker News post links an Anthropic page inviting people to share their thoughts on AI.