Researchers used Claude to hack OpenAI

Security researchers, identified as three Indian individuals, employed Anthropic’s Claude model to infiltrate OpenAI’s systems, taking over staff accounts and accessing a private software cache. India Today ran a profile of the three Indians, while ABC News released a video highlighting the discovery of another AI vulnerability. The Washington Post framed the incident as a warning that the AI industry faces a broader security problem. Gizmodo emphasized the rapid timeline, stating the hack was achieved in under 72 hours. The Hacker News detailed that Claude Opus 5 was used to exploit chained flaws and seize OpenAI staff accounts, a point echoed by The Verge in its coverage of the same technique.

PYMNTS.com focused on the breach of OpenAI’s private software cache, describing the attack as a penetration of internal infrastructure rather than a superficial exploit. Semafor expanded the narrative to suggest that AI cybersecurity risks are exploding as tools like Claude become weaponized against rivals. CBS News reported that AI security experts themselves confirmed using Claude to hack ChatGPT, underscoring the cross‑model threat landscape. Collectively, the outlets underline both the technical novelty of using one AI to attack another and the urgent need for stronger safeguards. The incident arrives amid heightened scrutiny of AI safety, where model alignment has dominated discourse but infrastructure security has received less public attention.

Anthropic’s Claude, a competitor to OpenAI’s ChatGPT, now sits at the center of a debate about responsible AI deployment and the potential for inter‑model exploitation. Observers can expect official responses from OpenAI and Anthropic, possible patches to the identified flaws, and renewed regulatory interest in AI system hardening. Future coverage will likely track remediation efforts, any legal actions, and broader industry moves to address the highlighted security gaps.