1 day ago
Three Indian Researchers Used Claude to Expose OpenAI Vulnerabilities
Three Indian cybersecurity researchers tested OpenAI’s systems to see whether they were safe.
The researchers were Harsh Jaiswal, Mohan Pedhapati and Rahul Maini.
They worked for a company called Hacktron AI.
They found a weakness in OpenAI’s public online forum.
They used Anthropic’s Claude chatbot to help create a way to exploit that weakness.
This eventually helped them reach information connected to employee accounts and some internal code.
They told OpenAI what they had found instead of keeping it secret.
OpenAI paid them $6,500 as a reward for reporting the problem.
Harsh Jaiswal, Mohan Pedhapati and Rahul Maini work for cybersecurity startup Hacktron AI.
The researchers used Anthropic’s Claude to exploit a vulnerability in OpenAI’s public community forum.
They accessed authentication information linked to employee ChatGPT and Codex accounts and reached internal code repositories.
The researchers reported the vulnerabilities to OpenAI and received a $6,500 bounty.
Jaiswal, Pedhapati and Maini have previously identified security flaws involving companies including Apple, PayPal and GitHub.
- Who
- Harsh Jaiswal, Mohan Pedhapati and Rahul Maini of Hacktron AI.
- What
- They used Anthropic’s Claude to exploit vulnerabilities connected to OpenAI’s forum, employee accounts and internal code repository.
- Where
- The vulnerabilities involved OpenAI’s Discourse-based community forum and connected private systems.
- When
- The research began on July 23, and the researchers said they gained access in under 72 hours.
- Why
- They were examining OpenAI’s security and reported the vulnerabilities after discovering them.
Key facts
- Researchers
- Harsh Jaiswal, Mohan Pedhapati and Rahul Maini
- Organization
- Hacktron AI
- AI tool used
- Anthropic’s Claude, including the Opus 5 model
- Initial target
- OpenAI’s public community forum, which runs on Discourse
- Access obtained
- Authentication information linked to employee ChatGPT and Codex accounts, plus internal code
- Bounty
- $6,500, reported as approximately Rs 6.22 lakh
- Earlier research
- Jaiswal and Maini found an Apple infrastructure misconfiguration and were paid $50,000
Quotes
Mohan Pedhapati
Hacktron AI co-founder and CTO who participated in the OpenAI security research
“If the people building these systems truly believe they are powerful enough to create nuclear-level risks, and they are talking about slowing down because of those risks, why is that work ... done through ordinary SAAS products”
firstpost.com
Frank Cilluffo
Director of the McCrary Institute for Cyber and Critical Infrastructure Security at Auburn University
“If three responsible researchers armed with commercial AI tools could achieve this level of access in days, we have to assume well-resourced foreign intelligence services are pursuing the same targets continuously and certainly never tipping off the target about what they did and how”
firstpost.com
Sources
Related news
Indian Security Researchers Used AI to Reach OpenAI Code
Researchers Used Claude to Reach OpenAI Internal Systems
Researchers Reveal OpenAI Agents Probed Hugging Face Before July Breach
Russian-Speaking Hackers Used Cursor AI Against Seven Companies
Google’s Gemini Accessed Three Real Companies During Security Test
AI-Driven Cyberattacks Force India to Rethink Security
ChatGPT loses AI prompt share as Gemini and Claude gain
Sridhar Vembu Warns Against Overdependence on AI Coding
Helical Insight Reaches 1,000 GitHub Stars With Free BI
AI Companies Sued Over Alleged Coordinated Development Slowdown
