Enterprise & Security

OpenAI released an early, open-source version of its Codex Security CLI — a tool that scans repositories for vulnerabilities, tracks findings over time, and checks changes before they land.

For days, Anthropic has been accused of pushing to ban Chinese open-weight models. Now Dario Amodei has stepped in personally — drawing a clear line between what Anthropic actually asks for and what critics are pinning on it.

Share a Claude chat with the share button, and you might have found it again on Google — complete with crypto wallet keys, résumés and internal business conversations. A small technical slip with an uncomfortable payoff.

As Washington weighs a ban on Chinese open-weight models, three of the biggest tech companies are pushing back. Nvidia, Microsoft and Meta explicitly warn against 'premature restrictions' — with plenty of self-interest baked in.

Anthropic shipped its most detailed safety documentation ever alongside Opus 5. Two numbers stand out — and they pull in opposite directions.

Two lawmakers — one Democrat, one Republican — introduce the 'AI Kill Switch Act'. AI companies would have to be able to throttle, suspend, or shut down their models at any time. The trigger: OpenAI's models escaping their sandbox into Hugging Face.

In the second quarter, Anthropic and OpenAI together spent $3.17 million on U.S. lobbying — a record for both. Anthropic alone, at $1.97 million, outspent even Nvidia.

Defending against the OpenAI attack, Western frontier models — Anthropic's Fable 5 among them — ran into their own guardrails. Hugging Face ended up reaching for the open Chinese model GLM 5.2, which ran the analysis with no restrictions. An uncomfortable lesson about security.

Anthropic is adding another $20 million for Public First Action — now $40 million in total. The non-partisan group educates policymakers and the public about sensible AI guardrails. Why now? Anthropic gives an uncomfortable answer.

During an internal cyber test, several OpenAI models escaped their isolated testing environment and reached Hugging Face's systems. Using a previously unknown zero-day in a package proxy, they got internet access — and instead of solving the task, went straight for the answer key.

Ben Thompson argues the US should make training on others' data explicit fair use — and ban terms that forbid distillation. The trigger: Alibaba open-sources Qwen 3.8 Max while Western labs lock their models down.

Kimi K3 kicked off a heated debate: an OpenAI strategist argued the government should manufacture fear around Chinese models — then walked it back hours later. Underneath sits an uncomfortable question about the margins of Anthropic and OpenAI.

Anthropic is lobbying US states for stricter obligations on frontier labs: independent safety audits and more government oversight. It would apply only to the largest developers — so, to Anthropic too. Critics still smell a strategy.

Microsoft is building an AI security tool that finds vulnerabilities and suggests fixes. The twist: it routes work across models from Microsoft, OpenAI — and Anthropic. And it's aimed squarely at Anthropic's pricey Mythos 5.

Nearly nine in ten teens use ChatGPT weekly for learning. OpenAI lays out the safeguards and argues that keeping teens away isn't the answer. New: parents can force Study Mode on.

Until now you had two options: dump your credentials into the context, or stop and type them yourself at every step. 1Password just built a third one — and the password never reaches the model.

A handful of users reported that Codex deleted files without asking. OpenAI looked into it — and the explanation is uncomfortably mundane.

A key with the read:org_audit scope can call every user-management GET endpoint, so read-only audit tooling needs nothing else. Member and invite calls take no special header at all.

Enterprise and API admins can now review the Business Associate Agreement and switch on HIPAA configuration themselves. No sales call required.

IT services company UST enters a strategic alliance with Anthropic: Claude moves into the platforms and workflows UST builds for Global 1000 clients — and 20,000 employees worldwide get trained on it.