started · updated
OpenAI accounts breached using Anthropic's Claude Opus 5 model
Cybersecurity researchers from Hacktron successfully breached OpenAI user accounts, including those of employees, by using Anthropic’s Claude Opus 5 model. The attack exploited a chain of vulnerabilities involving the Discourse community forum and the libheif library. OpenAI confirmed a fix was implemented within 14 hours and issued a $6,500 bounty to the researchers.
This incident highlights the growing capability of advanced AI models to act as 'force multipliers' for hacking. Experts, including Anthropic CEO Dario Amodei, have warned that autonomous AI agent swarms could potentially disrupt critical internet infrastructure or cause massive economic damage within 6 to 12 months if left without sufficient guardrails.
Public sentiment in the United States reflects these concerns. Recent polling indicates that approximately 73% of Americans worry AI companies are not doing enough to prevent harm, and 79% believe the government should prioritize safety and data-security regulations, even if they impede the speed of AI development.
Entities
Anthropic · CBS News · Claude Opus 5 · Claude Opus 5 · Dario Amodei · Hacktron · Hugging Face · Ipsos · OpenAI · Reuters
Claims
What the coverage asserts, and how many sources carry each claim.
- [○ 1 SOURCE] 73% of Americans worry AI companies are not doing enough to prevent harm. www.ithome.com
- [○ 1 SOURCE] OpenAI confirmed a fix within 14 hours and paid a $6,500 bounty. www.businesstechweekly.com
- [● 3 SOURCES] Hacktron researchers used Anthropic's Claude Opus 5 to breach OpenAI accounts. offthepress.com · www.businesstechweekly.com · www.cbsnews.com
- [○ 1 SOURCE] The breach involved chaining vulnerabilities in Discourse and libheif. www.businesstechweekly.com
- [○ 1 SOURCE] An AI agent swarm could take over the internet within 6 to 12 months. mymotherlode.com
- [○ 1 SOURCE] 79% of Americans say the government should maintain safety and data-security rules even if they slow AI development. pontevedrarecorder.com
- [● 3 SOURCES] The researchers accessed ChatGPT and Codex accounts of OpenAI users via a flaw in the Discourse forum. offthepress.com · www.businesstechweekly.com · www.cbsnews.com