Anthropic’s New AI Can Design Hardware, Find Cybersecurity Flaws and Solve Physics Problems

Anthropic on Monday introduced updated versions of its Claude Fable and Claude Mythos models, touting advances in coding, cybersecurity, hardware engineering and scientific research as the company pushes AI into more complex technical work.

During pre-release testing, the models helped engineers design three working electronic devices, identify software vulnerabilities, solve two open problems involving black holes and formalize a proof of Fermat’s Last Theorem in just 30 hours, according to a blog post from Anthropic.

Claude Fable 5.2 and Claude Mythos 5.2 generate outputs up to 30% faster than their predecessors and come with lower pricing of $8 per million input tokens and $40 per million output tokens. An optional fast mode can increase token-generation speeds by up to 2.5 times.

AI Moves Beyond Coding Into Hardware Design

Anthropic highlighted the models’ ability to work beyond traditional software development, including designing physical hardware.

In one experiment, the company gave Fable 5.2 a single-paragraph prompt to design three electronic devices: a home voice-assistant hub, a USB speakerphone and a handheld virtual pet.

The model selected electronic components, designed circuit boards, created 3D-printed enclosures and programmed the devices. Anthropic said all three devices worked without requiring electrical modifications, with each circuit board ready for manufacturing in less than a day.

Cybersecurity and Scientific Breakthroughs

Anthropic described Mythos 5.2 as its most capable model for cybersecurity engineering, highlighting its ability to identify, reproduce and fix software vulnerabilities.

In an evaluation involving 1,829 known vulnerabilities, Mythos 5.2 identified 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1. One security company also found that one in six confirmed bugs discovered by Mythos 5.2 had been missed by every other frontier model it tested.

The cybersecurity capabilities are restricted to vetted organizations through Anthropic’s Cyber Verification Program, while Fable includes safeguards that block most high-risk cybersecurity and biological research tasks.

The models also made progress in scientific research. Anthropic said Fable 5.2 completed a calculation physicists had outlined in 2003 but had not been able to finish, resolving a question involving black hole formation in small theoretical universes.

In mathematics, a team of 1,000 Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in 30 hours, compared with 11 days for an earlier effort using an unreleased Claude research model. Anthropic said the latest effort required one-third as many action tokens and produced a proof seven times shorter.

Claude Takes On More Engineering Work

Anthropic also pointed to growing reliance on Claude in its internal software development.

More than 90% of changes merged into the company’s core codebase are now written by Claude, with human guidance. In one example, AI agents migrated a critical production service from Python to Rust, making it more efficient and reliable while improving startup speed by approximately 26 times.

Early users also reported improvements in the model’s ability to understand instructions and complete complicated projects.

"The model is in my head. It understands the intent behind every prompt I’ve given it. I haven’t had to say ‘No, I meant…’ a single time," said Joe Haddad, a member of Vercel’s technical staff.

Anthropic said Fable 5.2 is available broadly, while Mythos 5.2 remains restricted to approved organizations through its cybersecurity and life sciences programs.

Photo: Below the Sky / Shutterstock