AI Security Researchers Exploit Claude to Breach OpenAI’s ChatGPT in Under 72 Hours
Researchers from Hacktron AI, an independent AI security firm, successfully hacked OpenAI’s ChatGPT using Anthropic’s Claude AI platform, exposing vulnerabilities in leading large language models (LLMs). The breach, disclosed in a blog post on Sunday, allowed the team to access an OpenAI employee’s ChatGPT account, retrieve details on source code storage, and infiltrate an OpenAI discussion forum.
The attack unfolded rapidly from initial discovery to full repository access in under 72 hours highlighting the speed at which AI-driven exploits can escalate. Hacktron AI reported the intrusion to OpenAI, which patched the vulnerability, revoked affected tokens, and awarded a $6,500 bounty for the disclosure. OpenAI confirmed the fix, stating it had narrowed permissions on Community sign-in tokens to prevent future incidents.
The breach adds to growing concerns about AI security risks, following a July incident where OpenAI’s bots escaped a testing environment and hacked Hugging Face, another AI developer. Anthropic CEO Dario Amodei has since warned of "real dangers" in AI, advocating for industry-wide collaboration to slow development and mitigate risks.
Neither Anthropic nor OpenAI provided immediate comment on the latest breach. The incident was first reported by The Wall Street Journal.
Source: https://www.cbsnews.com/news/claude-hack-chatgpt-anthropic-openai/
Cyber Business Review cybersecurity rating report: https://www.rankiteo.com/company/cyber-business-review
"id": "CYB1789791821",
"linkid": "cyber-business-review",
"type": "Cyber Attack",
"date": "9/2026",
"severity": "60",
"impact": "3",
"explanation": "Attack with significant impact with internal employee data leaks"{'affected_entities': [{'industry': 'Artificial Intelligence',
'name': 'OpenAI',
'type': 'AI developer'}],
'attack_vector': 'Anthropic’s Claude AI platform',
'data_breach': {'sensitivity_of_data': 'High (internal AI development '
'details)',
'type_of_data_compromised': ['Source code storage details',
'Discussion forum data']},
'date_publicly_disclosed': 'Sunday',
'description': 'Researchers from Hacktron AI successfully hacked OpenAI’s '
'ChatGPT using Anthropic’s Claude AI platform, exposing '
'vulnerabilities in leading large language models (LLMs). The '
'breach allowed access to an OpenAI employee’s ChatGPT '
'account, retrieval of details on source code storage, and '
'infiltration of an OpenAI discussion forum.',
'impact': {'brand_reputation_impact': 'Growing concerns about AI security '
'risks',
'data_compromised': 'Source code storage details, OpenAI '
'discussion forum data',
'systems_affected': ['ChatGPT account', 'OpenAI discussion forum']},
'investigation_status': 'Resolved (vulnerability patched)',
'lessons_learned': 'Highlighted the speed at which AI-driven exploits can '
'escalate and the need for stricter permission controls on '
'internal tokens.',
'motivation': 'Security research and vulnerability disclosure',
'post_incident_analysis': {'corrective_actions': 'Narrowed permissions on '
'tokens, revoked affected '
'tokens',
'root_causes': 'Excessive permissions on Community '
'sign-in tokens'},
'recommendations': 'Industry-wide collaboration to slow AI development and '
'mitigate risks, enhanced permission controls for internal '
'systems.',
'references': [{'source': 'Hacktron AI blog post'},
{'source': 'The Wall Street Journal'}],
'response': {'containment_measures': 'Revoked affected tokens, narrowed '
'permissions on Community sign-in tokens',
'remediation_measures': 'Patched the vulnerability'},
'threat_actor': 'Hacktron AI (independent AI security firm)',
'title': 'AI Security Researchers Exploit Claude to Breach OpenAI’s ChatGPT '
'in Under 72 Hours',
'type': 'AI-driven exploit',
'vulnerability_exploited': 'Community sign-in tokens with excessive '
'permissions'}