← Back to the wire

Researchers used Anthropic's Claude to hack into OpenAI

AchievementResearchSep 18, 2026

Security startup Hacktron AI used Anthropic's Claude Opus 5 to breach OpenAI through its bug-bounty program, earning a $6,500 award, The Wall Street Journal reported. The team chained two vulnerabilities — a memory bug in the libheif library behind OpenAI's Discourse forum and a second flaw — to access employee ChatGPT and Codex accounts. OpenAI says the issues are resolved. Hacktron noted Opus 4.8 failed the exploit until Opus 5's release succeeded within hours.

Receipt № 19551 source · awaiting confirmation ◐

Evidence

1source· awaiting independent confirmation

No score is assigned. Sources and their independence are shown in the citation chain below.

Citation chain · 1 source

OpenAICompanyChatGPTModelHugging FaceCompanyAnthropicCompanyClaudeModelThe Wall Street JournalCompanyGray SwanModelMatt FredriksonPersonOpus 5ModelHacktron AICompany
Canonical: https://techcrunch.com/2026/09/18/researchers-used-anthropics-claude-to-hack-into-openai/