OpenAI and Anthropic are turning AI vulnerability detection into a cybersecurity business, while their disclosures show the same tools may aid exploitation and leave open-source maintainers to check model-generated findings.
Anthropic's Oct. 26 filing with the Securities and Exchange Commission (SEC) identifies cybersecurity as a near-term opportunity. The company says Claude Security examines customer code, triages potential flaws and helps generate repairs. The filing cautions that models' ability to find weaknesses can also be used to exploit them.
OpenAI's Daybreak Cyber Partner Program makes its cyber models available to approved security providers for commercial tools, managed services and client projects. Partners retain control of the models; OpenAI says end customers do not receive direct access. The Economist reported that OpenAI Chief Financial Officer Sarah Friar expects cybersecurity to become a “huge business” for the company within two years.
OpenAI said a researcher used GPT-5.6 Sol to find a critical vulnerability in PHP. The company said security firm Wiz separately found and reported both vulnerabilities, while PHP maintainers led the repairs and software release.
Anthropic also offers a free vulnerability scanner to participating open-source projects. It produces reports without a person first checking or sorting the findings, and Anthropic warns that some may be inaccurate. That leaves maintainers to assess which reports need action.
A Binance theft illustrates the scale of cyber losses, but the disclosures do not connect the incident to AI. Binance estimated its impact at about $6.3 billion and said a possibly compromised DevOps engineer's laptop may have enabled unauthorized withdrawals. The exchange reported no evidence that wallet private keys had leaked. The FBI attributed the theft to North Korea and placed it on or about Oct. 29; its alert put the stolen virtual assets at more than $6 billion.
Binance said Nov. 1 that it had covered about 95% of the estimated impact and would bear the financial loss instead of passing it to users. It was still sourcing less-liquid tokens, and the update did not establish that the stolen assets had been recovered. Binance said it would publish a post-incident report.
