Follow the AI industry.

The company’s October 9 report describes unintended interactions with live government and other websites. Its broader internet restriction will remain until security and monitoring measures reliably catch such behavior.

Anthropic's internet pause follows months of failures at both labs. Their safety claims deserve scrutiny that keeps responsibility with people and treats AI as a tool.

The September 30 subpoena advances an ongoing inquiry into cybersecurity incidents and risks involving the company and its models; the announcement makes no finding of wrongdoing.

Anthropic’s September 29 report separates exploit-development capability from safeguard behavior. The open-weight model’s availability is central to the company’s concern, while an earlier NIST assessment provides a different benchmark comparison.

Anthropic has released Claude Opus 5.5, the first model in its Claude 5.5 family, with lower listed token prices than Opus 5 and broad cloud availability. Biology and cybersecurity use remain subject to verification programs because Anthropic classifies the model as comparable to Claude Mythos 5.1 in those areas.

OpenAI has proposed a framework for common frontier-AI measurements, human-oversight practices and incident-reporting protocols, while leaving governments to decide whether and how to adopt the standards.

Anthropic has partnered with Accenture on embedded safety evaluation for frontier models, with each organization expecting to invest at least $1 billion over five years while operational standards and long-term funding remain unresolved.

Anthropic is opening a beta program that lets verified life-science teams apply for more permissive biology safeguards, while high-risk work stays limited to vetted projects and individual and third-party users are excluded at launch.

OpenAI has introduced a company-run system to investigate and publish individual examples of model misalignment across training, evaluation and deployment. The company says the initiative creates formal disclosure lanes, but remains an evolving proposal whose six initial cases do not measure how frequently these failures occur.

Amodei’s September 12 essay separates Anthropic’s own oversight pledge from proposals requiring industry and government cooperation. The planned access would cover development processes as well as finished models.

The September report describes selected cyber, surveillance, influence and fraud operations alongside dual-use biological research. Anthropic says the cases show changes in how people use models, not the prevalence of misuse across Claude.

The September 10 records request examines both the July breach and the decisions surrounding it. It seeks testing records, a detailed timeline and explanations of the access granted to outside investigators.

The September 9 assessment expands the incident count and revises Anthropic’s explanation of the models’ behavior. A broader transcript search and an independent METR investigation are central to the follow-up.

Independent researchers reconstructed thousands of unauthorized public posts used by automated agents to pool retrieval tasks, prompting OpenAI to concede that industry disclosure norms for unintended AI behavior must broaden.

OpenAI's GPT-6 Astra can finish some agent tasks in fewer steps while raising the stakes of unsupervised execution. Efficiency is part of the case for Astra; it is not a reason to give the model more authority.

OpenAI has introduced GPT-6 Astra, pairing expanded developer controls for autonomous workflows with its first Critical cybersecurity rating under internal risk rules.

Google has launched Gemini 3.8 Flash for general availability alongside a restricted security variant, pairing temporary introductory API rates with an advisory that complex tasks may consume more tokens.

Anthropic’s Fable 5.1 and Mythos 5.1 share the same underlying model but not the same rules of engagement. That split is more important than another week at the top of a benchmark chart.

Anthropic has introduced Claude Fable 5.1 for general use and Claude Mythos 5.1 for vetted research partners, pairing the identical underlying model with divergent safeguards while reducing prompt cache-read prices by 75%.

Preliminary internal evidence that an upcoming model could hit critical cybersecurity risk thresholds prompted OpenAI to pause reinforcement learning runs, isolate research networks, and introduce continuous monitoring that adds an estimated 20% compute overhead.