Startup Hacks OpenAI Using Anthropic's Claude AI

Startup Hacks OpenAI Using Anthropic's Claude AI

Sam Altman, OpenAI's chief executive, in a hedcut engraving portrait seated at a desk with hand raised.
Sam Altman, OpenAI's chief executive, whose systems were probed by a San Francisco cybersecurity startup using Anthropic's Claude.Illustration: The Frank

The News

A tiny San Francisco cybersecurity startup used a rival company's AI to break into OpenAI's systems — and got paid for it.

Hacktron flagged vulnerabilities in OpenAI's systems and was paid $6,500 for the discovery, Business Insider reported. Fortune reported that three researchers using Anthropic's Claude got into OpenAI and accessed its source code, collecting the same $6,500 reward.

The payment was a bug bounty — money companies offer outside researchers who report security holes instead of exploiting them.

Timeline

Sept. 18, 2026

Forbes reports that security researchers hacked into OpenAI using Anthropic's Claude, and that OpenAI paid them a $6,500 bug bounty.

Sept. 18, 2026

Business Insider reports the researchers work for Hacktron, a San Francisco-based startup, which flagged vulnerabilities in OpenAI's systems and was paid $6,500 for the find.

Sept. 18, 2026

Fortune reports the team was three people using Anthropic's Claude, and that they accessed OpenAI's source code.

Reactions

Forbes framed the episode as a warning shot, writing that the incident "once again highlights the growing cybersecurity threats posed by advanced AI agents."

OpenAI's response was the bounty itself: it paid the researchers $6,500 under its bug bounty program rather than treating the intrusion as an attack.

What's Next

The open question is how much cheaper AI has made this kind of work. A team of three, using a commercially available AI model, found holes in one of the best-funded software companies on earth for a payout of $6,500.

Expect more scrutiny of what AI agents can do in the hands of researchers — and in the hands of people who are not filing bug reports.

Poll

Was a $6,500 bounty the right price for finding holes in OpenAI?

Too low — pay researchers far more
0.0%
About right
0.0%
Bounties shouldn't exist at all
0.0%
Not sure
0.0%

Trump Weighs 12 Data Centers on Public Lands

Spanish Police Clear Migrants From Ceuta Beaches

Suicide Car Bomb Kills 16 at Pakistan Mosque

Buffett Exits Berkshire Chair; Son Howard Takes Over

NHS Trust Admits 20 Unnecessary Mastectomies, Hundreds Harmed

37 Die in Nigerian Custody; Police Tear-Gas Protest

Video Shows NBC Helicopter Plunge That Killed Three

Putin Seizes Nestlé's Russian Business by Decree

Russians Vote in Election Putin Cannot Lose

Early Voting Opens With Congress Up for Grabs

South Korea Rejects Trump's Call for Iran Troops

Malaysia's King Moves Ex-PM Najib to House Arrest

Anthropic Says Claude Now Helps Build Its Successor

Japan Hikes Rates to 31-Year High

Streisand Blasts Trump's 'Unchecked Ego' Over Kennedy Center

Pentagon Resumes Testosterone Testing for Troops Over 30

Trump Signs Orders Opening Federal Land to Hunters

Report Blames Haitian Police for 164-Death Massacre

Trump Plans First Meeting With Venezuela's Leader

Microsoft Exec Called AI Scraping History's Largest Theft