GLM 5.2 evaded 70 percent of the time and Claude Opus 5 just 2 percent, across 50 tasks. If you trust a monitor to stop your agents, test it with a task that pushes against it.
Partners include CrowdStrike, Cursor, Snowflake and Harvey, plus a consulting network. MSPs can list their own connectors or services through the partner program.
In LangChain's test, a tuned Kimi K3 scored 96 against 90 for the base model. If you log agent traces already, this is a path to cheaper, faster agents.
OpenAI opened its Agents API in public beta this month, exposing the harness that powers Codex with managed sessions, tool The post OpenAI and Cursor… · The New Stack
The ask came from the Office of the National Cyber Director, and Anthropic has not given Mythos 5.1 to the UK institute. Allied safety testing may now wait on Washington.
It can grow by another $9 billion, and Anthropic gets a warrant for up to 5 percent of Akamai. A content network company is now an AI compute supplier.
He told NBC News each lab should manage safety itself, the same day Amodei and Altman briefed the UN Security Council. Meta's Muse agent became the top free app after its September 8 launch.
Two sources told The Washington Sun about the classified estimates; the exact figure is not public. A budget office estimate for one testing bill was about $20 million a year.
The Project Suncatcher prototype, built with Planet, flies on SpaceX's Transporter-18. Google says future satellites would each carry dozens of TPU chips.
The BriefA federal appeals court in Washington ruled 2-1 on Friday that the Pentagon may keep Anthropic's Claude out of its supply chain, which means the military and its contractors cannot use Claude in their Defense Department work. The ban began in March after Anthropic refused to drop two limits in its contract: no fully autonomous weapons and no mass surveillance of Americans. If you support any company that does defense work, check which AI tools touch that work this week, because a California ruling in August against a different label does not undo this one.
Level UpRead the short federal law behind the ban, 41 U.S.C. 4713, before a defense client asks you about it. It lets an agency bar a supplier from its own contracts and from any subcontract that uses that supplier for the agency's work. Then list every client that does defense work and which AI tools, including Claude inside other products, touch that work. →41 U.S.C. 4713, the supply chain law behind the Claude ban