The U.S. Cybersecurity and Infrastructure Security Agency (CISA) is using Anthropic's AI model Mythos to audit government software for potential security flaws, three people familiar with the matter said. The initiative aims to scan government code repositories for bugs that could be exploited by foreign spies and cybercriminals, highlighting the agency's growing interest in leveraging advanced AI tools for cybersecurity defense.

CISA's Attack Surface Evaluation team is conducting the scanning, according to one source. This unit specializes in digital security assessments and hacking exercises across government infrastructure. Two sources indicated that the audits have already uncovered a large number of vulnerabilities, though specific details regarding the nature, severity, or volume of code scanned were not disclosed.

The reported deployment comes after a turbulent period in Anthropic's relationship with the U.S. government. Tensions escalated earlier this year after the San Francisco-based AI startup reportedly refused to remove safeguards designed to prevent its models from being used for autonomous weapons or domestic surveillance. The dispute led the Pentagon to designate Anthropic as a supply-chain risk, although a federal judge later blocked that designation. Relations have since improved following the private release of Mythos, an AI model reportedly built with advanced cybersecurity capabilities.

Government interest in Mythos appears to extend beyond CISA. The National Security Agency (NSA) has also utilized Mythos, despite existing government blacklists. Previous reports said NSA analysts tested Mythos in classified environments and were impressed with its cybersecurity performance. Anthropic later released a public version of the model called Fable with cybersecurity safeguards. The White House reportedly pushed the company to block foreign access to the model, triggering a temporary global shutdown that was lifted last week.

Anthropic did not respond to inquiries about the specific initiative with CISA. A representative for CISA previously stated they would look into the matter but has not provided further comments. The adoption of AI models for critical security audits underscores a shift in federal cybersecurity strategies, aiming to proactively identify and mitigate risks in government software systems.

Jim Cramer has urged investors to exercise caution regarding reports that Anthropic is in discussions with Samsung Electronics Co. to manufacture a custom AI chip, emphasizing that the news remains unconfirmed by either party. The potential partnership aims to leverage Samsung's manufacturing capabilities to produce hardware tailored for Anthropic's AI models, though no final agreement has been announced.

Anthropic is reportedly in discussions with Samsung about developing a custom AI chip. The project appears to be in its early stages, reported TechCrunch. As per the report, Anthropic has not yet determined the chip’s exact purpose, how it would integrate into its server infrastructure or the level of performance it is targeting. The company has also not made a final decision on whether it will proceed with the effort.

The reported talks follow an April report that said Anthropic had been considering designing its own AI chips to help address ongoing supply constraints and reduce reliance on third-party hardware. The move, if it materializes, would place Anthropic among a growing list of AI companies pursuing custom silicon to power their models more efficiently while reducing dependence on Nvidia Corp, which continues to dominate the AI chip market.

AI Chip Race Intensifies

The report also comes shortly after rival OpenAI announced a partnership with Broadcom Inc. to develop its own inference processor, known as “Jalapeño.” OpenAI has said the chip delivers improved performance per watt compared with competing offerings. Meanwhile, Amazon.com, Inc. and Alphabet Inc.’s Google already deploy proprietary AI chips across their cloud platforms. Samsung is a major player in semiconductor manufacturing and works closely with Nvidia while also collaborating with other technology companies on AI chip development.

Market Reaction and Company Stance

Reacting to the report, CNBC’s Cramer questioned whether investors should treat the news as credible before either company confirms it. "Now, Anthropic, the most promotional private company I may have ever seen, is said to be building its own custom chip with Samsung," Cramer said. He added that reports based solely on unnamed sources can move markets prematurely. "Of course that sends everything down because we really don't have anything but a rumor. I want to hear it from either company or both."

Neither Anthropic nor Samsung has publicly announced a custom AI chip partnership. However, in a statement to TechCrunch, Anthropic said its compute strategy will continue to depend on a diversified hardware stack featuring chips from Google, Amazon and Nvidia. It declined to comment further on the reported collaboration with Samsung.