LowNewsLLM-specific

Researchers used Anthropic’s Claude to hack into OpenAI

Source: TechCrunch (Security)(opens in a new tab)
Published: September 18, 2026
Record updated: October 9, 2026

Summary

Independent researchers at Hacktron AI used Anthropic's Claude to chain two vulnerabilities and gain access to multiple OpenAI employee ChatGPT accounts, as part of an OpenAI bug-bounty program, earning a $6,500 award. The entry point was a memory bug in libheif, reached through a HEIF/HEIC image upload processed by ImageMagick in Discourse, the software powering OpenAI's community forum. Hacktron said Claude Opus 4.8 could not build a working exploit, but Opus 5 succeeded within hours of its release.

Mitigation

Discourse issued a fix on July 27, and OpenAI says it has resolved the issues Hacktron uncovered. The libheif bug had already been fixed by its developers months earlier, but that fix was never flagged as a vulnerability and received no CVE.

Details

Scores and exploitation

Disclosure date: September 18, 2026

Affected

Vendors: OpenAI,
Anthropic

Classification

Attack type: Other
Attack sophistication: Moderate
AI component targeted: API
Impact: Confidentiality,
Integrity
Labels: security,
industry

How this record was produced

Summaries, mitigations and labels on this page are generated automatically from the source. The linked source is authoritative.

Original source: https://techcrunch.com/2026/09/18/researchers-used-anthropics-claude-to-hack-into-openai/

First tracked: September 18, 2026 at 02:00 PM ET, from TechCrunch (Security)

Classifier: claude-haiku-5-5, prompt v4

Classifier confidence: 85% (self-assessed)

Corrections

2026-10-10: Severity changed from high to medium. Severity capped at medium. High and critical measure technical exploitability and are reserved for vulnerabilities and incidents; this record is a news report.

All corrections

Summary

Mitigation

Related items