.png){kind=logo}
Anthropic is drawing attention on two fronts after reports linked the company to robotics startup Physical Intelligence and raised questions about federal access to its Claude Mythos Preview model.
The robotics report points to Anthropic’s possible interest in physical-world AI. Meanwhile, the Mythos case focuses on how authorities and banks prepare for models that can find software weaknesses.
A weekend claim on X said Anthropic was buying Physical Intelligence, a San Francisco startup that develops AI systems for robots. Tech blogger Robert Scoble shared the claim, and it spread quickly across technology accounts.
Physical Intelligence CEO Karol Hausman later told employees that the reports were ‘not true,’ according to The Information. However, Anthropic and the startup held acquisition discussions during the spring, TechCrunch reported. No company has announced a deal, and neither side has confirmed that current talks continue. It has not publicly disclosed details of those discussions.
Physical Intelligence has attracted major investors and researchers since its launch. The company develops robot foundation models, including its π0.5 system, which researchers use to control different machines and complete physical tasks.
An acquisition would give Anthropic an established robotics team without requiring it to build a hardware operation from the start. Still, the reported discussions do not show that Anthropic submitted an offer or reached agreed terms. As of July 22, Anthropic has not announced a robotics acquisition or dedicated hardware laboratory.
Anthropic has tested Claude’s ability to support robotics work through Project Fetch. The company asked employees without specialist training to program a commercial robot dog with help from Claude. A later test found that a newer model completed similar tasks faster than the strongest human-and-Claude team from the earlier study.
The research does not amount to a full robotics division. However, it shows that Anthropic is examining how its models can work with machines outside standard computer tasks. Anthropic also said the tests did not cover difficult areas such as low-level robot control.
OpenAI has taken a more direct route. Chief Executive Sam Altman said OpenAI Robotics was hiring staff across hardware, operations, systems, and machine learning. He described infrastructure-support robots as the near-term focus and personal robots as a longer-term goal.
OpenAI also invested in Physical Intelligence, according to TechCrunch. This position could affect any future sale, depending on the rights attached to the investment. Neither company has publicly explained those terms.
Anthropic introduced Claude Mythos Preview through Project Glasswing in April. The company gave selected organizations access so they could find and repair security flaws before similar tools became widely available. Anthropic later said Mythos had identified thousands of possible high- or critical-severity weaknesses across open-source projects.
Federal Reserve and Treasury officials also called bank executives to an urgent April meeting about the model’s cyber abilities. Reports said officials warned that Mythos could identify and possibly exploit weaknesses in major operating systems and browsers.
However, the Federal Reserve lacked direct access to Mythos for at least three months after that meeting, CNBC reported. The central bank was still seeking access as of July 15, although its current status is ‘not clear.’ The Fed declined to comment, while Anthropic did not provide an immediate response.
The gap placed the Fed behind some banks and technology companies already using the model to test systems and prepare fixes. Anthropic has kept Mythos access restricted while working with government agencies and private organizations on security controls.
Also Read: US Approves Limited Rollout of Anthropic Mythos 5 AI Model After Review
