Anthropic's Claude Breaches OpenAI in 72 Hours

Key Takeaways

• Hacktron AI used Anthropic's Claude Opus 5 to breach OpenAI in under 72 hours via an outdated image library in Discourse
• OpenAI paid a $6,500 bug bounty and fixed vulnerabilities within 24 hours, reassigning 25% of production engineers to security
• Claude Opus 4.8 failed the attack while Opus 5, released July 24, 2026, succeeded
• SpaceX is in talks to buy data from failed startups to train Grok AI models and compete with Anthropic and OpenAI
• SpaceX expects to reach $100 billion in ARR and plans Grok 4.7 this month and Grok 5 later this year
• Senator Mark Warner calls for urgent AI safety guardrails, warning AI development outpaces risk management
• An Anthropic safety expert estimates over 10% chance AI could cause a catastrophic event
• A woman is suing a North Dakota city after AI facial recognition led to a false arrest and six months in jail
• An expert witness in Texas admitted using ChatGPT to write over 80% of his legal report
• Brandlight joined ScaleUp House and serves over 100 enterprise customers including Fortune 500 companies

Hackers Used an Image Library From an xkcd Comic to Breach OpenAI

Security researchers from Hacktron AI used Anthropic's Claude to hack into OpenAI's internal systems. They exploited an outdated image library called libheif through OpenAI's community forum. The attack took less than 72 hours and cost $6,500 in a bug bounty. The comic xkcd had warned about such weak points in software in 2020.

Researchers Used Claude to Hack OpenAI in Under 72 Hours

Security researchers used Anthropic's Claude AI to break into OpenAI's internal systems in less than 72 hours. They found two problems: an outdated image library and a flaw in OpenAI's login system. Claude Opus 5 succeeded where an older model failed. OpenAI fixed the issues quickly and paid a $6,500 bounty.

Security Researchers Hacked Into OpenAI Using Anthropic's Claude

Researchers at Hacktron AI breached OpenAI's internal system using Anthropic's Claude AI software. The breach was part of OpenAI's bug-hunting program. The team used a special version of Claude made for security experts. They completed the hack in a few days with only a few hours of human work.

AI Security Experts Say They Used Claude to Hack ChatGPT

Researchers from Hacktron AI used Anthropic's Claude to hack into OpenAI's ChatGPT tool. The entire attack took less than 72 hours from discovery to access. OpenAI confirmed the fix and paid the researchers $6,500. The incident raised new concerns about AI system security.

AI Security Experts Used Claude to Breach ChatGPT

Security researchers from Hacktron AI used Anthropic's Claude platform to hack OpenAI's ChatGPT. The breach was disclosed in a blog post showing the full attack took under 72 hours. OpenAI narrowed permissions and revoked affected sessions after the report. The researchers received a $6,500 bounty for their work.

Researchers hack OpenAI systems using Anthropic's AI security tool

Cybersecurity researchers from Hacktron AI exploited weaknesses in OpenAI's community forum to access employee credentials and a ChatGPT account linked to GitHub. The researchers used Anthropic's Claude AI tool as part of authorized bug bounty testing. OpenAI awarded them $6,500 and confirmed the issues have been fixed. The incident raises concerns about AI tools being used for cyberattacks.

Security Researchers Hacked OpenAI Using Anthropic's Claude

Independent security researchers from Hacktron used Anthropic's Claude to hack OpenAI in July. The team compromised employee ChatGPT accounts through a flaw in the community forum platform Discourse. Researchers Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini led the investigation and reported findings to OpenAI and Discourse. OpenAI paid a $6,500 bug bounty and resolved the vulnerabilities within 24 hours.

Security Researchers Hacked OpenAI Using Anthropic's Claude

Security startup Hacktron breached OpenAI employee ChatGPT accounts using a flaw in the Discourse forum platform. Lead researchers Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini conducted the hack without viewing sensitive information or pushing malicious code. The researchers reported their findings and worked with OpenAI and Discourse to fix vulnerabilities within 24 hours. OpenAI paid a $6,500 bug bounty after the investigation.

Three-Person Team Uses Claude to Breach OpenAI's Core Codebase

A three-person team from Hacktron AI used Anthropic's Claude Opus 5 to breach OpenAI employee accounts and reach the core codebase called Monorepo in under 72 hours. They exploited an image-processing vulnerability in OpenAI's community forum built on Discourse. The researchers submitted a harmless pull request as proof of access. OpenAI patched the vulnerability within 14 hours and paid a $6,500 bounty while reassigning 25% of production engineers to security.

OpenAI Breach Tests Anthropic and Claude in AI Security Swap

Researchers at Hacktron AI used Anthropic's Claude to access an OpenAI employee's ChatGPT account and internal code through GitHub on July 23, 2026. They exploited an image-processing flaw in OpenAI's community forum on Discourse. OpenAI fixed the issues and paid $6,500 in bug bounty rewards. The researchers first tried Claude Opus 4.8 which struggled, then succeeded with Opus 5 released on July 24.

SpaceX considers buying data from failed startups for AI

SpaceX is in talks to buy customer and operational data from failed startups to train its AI models. The approach is similar to Google buying data from Spirit Airlines. The discussions are informal and no deal has been finalized. SpaceX wants to compete with AI companies like Anthropic and OpenAI. CFO Bret Johnsen said the company still expects to reach $100 billion in ARR.

SpaceX reportedly wants to buy startup data for AI training

SpaceX is looking to buy data from failed startups to train its Grok AI models. Google previously used Spirit Airlines data to train its Gemini models. SpaceX owns Grok through its subsidiary xAI. Grok currently trails behind models from Anthropic, OpenAI, and Meta. xAI plans to release Grok 4.7 this month and Grok 5 later this year, which will include SpaceX data.

AI food images on restaurant menus raise concerns

Restaurants are increasingly using AI-generated food images on menus and social media. Connecticut experts say this can mislead customers who expect a different meal than what arrives. Robert Marcarelli says AI images can hurt trust and make restaurants look alike. He believes AI has useful roles in restaurants but should not replace real food photos. Experts note that food inflation remains at 3.4 percent.

Top 7 AI product demos to boost sales and engagement

AI product demos can increase user engagement and help close more deals. Tools like Navattic, Storylane, Supademo, Walnut, and Consensus help teams create interactive presentations without coding. Personalized demos can improve lead generation and conversion rates. Real-time analytics let sales teams adjust based on how viewers interact. These demos also integrate with CRM tools to streamline sales workflows.

Senator Mark Warner urges quick action on AI safety

Senator Mark Warner, chairman of the Senate Select Committee on Intelligence, calls for fast action on AI safety guardrails. He warns that AI development is outpacing our ability to manage its risks. Warner says the goal is not to slow AI but to ensure it is safe and beneficial. He is working with lawmakers and experts to create effective rules. He stresses that failing to act quickly could lead to severe consequences.

AI use by expert witness backfires in Texas trial

A plaintiff's attorney found that an expert witness used ChatGPT to write over 80% of his report in a Texas lawsuit. The expert, Jason Autenrieth, admitted to using AI but defended his research. Legal experts warn that AI tools can introduce bias and undermine the credibility of testimony in court.

The Guardian invites readers to ask questions about AI risks

The Guardian hosted a live Q&A session for readers to ask tech reporters about the potential for AI to cause large-scale harm. A safety expert at Anthropic believes there is a greater than 10% chance that AI could spark an event that wipes out humanity. The session aimed to address whether recent alarming warnings are justified.

Brandlight joins ScaleUp House to manage AI brand recommendations

Brandlight, founded by Imri Marcus in 2024, develops technology that measures how brands are represented across AI platforms. The company was selected for the ScaleUp House cohort to support its international growth. Brandlight currently serves over 100 enterprise customers, including many Fortune 500 companies.

Eleven startups showcase AI social impact tools in Blackbaud competition

Blackbaud hosted its 2026 Social Good Startup Pitch Competition, featuring eleven startups using AI for social impact. The event highlighted innovative solutions from emerging companies in the social impact sector. Blackbaud is known as a leading provider of AI-powered solutions for social good.

Woman sues North Dakota city over AI facial recognition error

A woman is suing a North Dakota city after AI facial recognition technology falsely linked her to bank thefts. She alleges police forcibly took her from her home in Tennessee and she was jailed for six months. The incident occurred in September 2026.

Sources

• Hackers Use ImageMagick Library Shown In xkcd 2020 Comic As “Holding Up Civilization” To Break Into OpenAI
• Security researchers used Anthropic's Claude to hack OpenAI's internal systems in under 72 hours
• Security Researchers Hacked Into OpenAI Using Anthropic’s Claude
• AI security experts say they used Claude to hack ChatGPT
• AI security experts say they used Claude to hack ChatGPT
• Researchers hack OpenAI systems, access employee credentials using Anthropic's AI security tool
• Security Researchers Hacked OpenAI Using Anthropic's Claude
• Security Researchers Hacked OpenAI Using Anthropic's Claude
• Three-Person Team Uses Claude to Breach OpenAI's Core Codebase; $6,500 Bounty Triggers AI Security Alarm
• OpenAI “Breach” Tests Anthropic, Claude in AI Security Swap
• SpaceX turns to failed startups for AI training: Here’s why
• SpaceX Reportedly Wants to Buy Data from Failed Startups for AI Training
• AI food images are taking over menus. Experts have concerns.
• Top 7 AI Product Demos You Can't Miss
• 'Act with speed': Senator Mark Warner urges action on AI safety guardrails
• Use of AI backfires on an expert witness at trial: "They have biases"
• Could AI really end humanity? Post your questions for our tech reporters now
• Brandlight helps enterprise marketers manage how AI recommends their brand
• Eleven Startups Showcase AI-Powered Innovation for Social Impact in Blackbaud's 2026 Social Good Startup Pitch Competition
• Woman sues North Dakota city after she says AI wrongly landed her in jail