A team of researchers called Hacktron AI used Anthropic's Claude Opus 5 to breach OpenAI's systems in under 72 hours during a bug bounty program. They exploited a vulnerability in the libheif image processing system, which also affects platforms like Slack, Meta, GitHub Enterprise, and Zoom. The attack cost less than $3,000 in AI tokens and resulted in OpenAI paying the researchers $6,500 for their findings.
Meanwhile, debate is intensifying around China's AI capabilities. Cohere CEO Aidan Gomez argues that China has developed world-class AI independently, pushing back against claims that its progress relies on stealing American models through distillation. He points out that Chinese models now outperform top U.S. models on certain benchmarks, aided by a large talent pool and strong government backing.
On the policy front, the IMF warns that AI could widen inequality across Europe despite boosting productivity by roughly 1 percent over five years. Around 60 percent of workers in advanced European economies face high exposure to AI, and the IMF recommends deeper EU integration and investment in domestic AI to reduce dependence on the U.S. and China.
Key Takeaways
- Hacktron AI used Claude Opus 5 to breach OpenAI systems in under 72 hours via a libheif vulnerability
- The hack cost less than $3,000 in AI tokens and affected Slack, Meta, GitHub Enterprise, and Zoom
- OpenAI paid researchers $6,500 through its bug bounty program
- Cohere CEO Aidan Gomez says China's AI progress is independent, not based on theft
- Chinese models now beat top American models on some benchmarks
- The IMF warns AI could widen inequality in Europe despite 1 percent productivity gains
- About 60 percent of workers in advanced European economies face high AI exposure
- Anthropic co-founder Jack Clark says liberal arts degrees will thrive in the AI era
- Atlassian turned Jira into an AI coding command center to monitor AI-driven development
- CoreWeave acquired Monolith AI to advance Physical AI in automotive and aerospace
Hackers Use Claude AI to Breach OpenAI Systems in 72 Hours
A team called Hacktron AI used Anthropic's Claude Opus 5 to breach OpenAI's security during a bug bounty program. They exploited a flaw in the Discourse forum platform and gained access to employee ChatGPT accounts and internal code repositories. The entire hack took less than 72 hours from discovery to full access. OpenAI patched the holes and paid the researchers $6,500. The incident shows how AI can lower the barrier for cyberattacks.
White Hat Hackers Breach OpenAI Using Claude in Under 72 Hours
Researchers in OpenAI's bug bounty program used Claude Opus 5 to exploit a vulnerability in the libheif image processing system. This allowed them to hijack an OpenAI employee's ChatGPT account connected to GitHub, gaining full repository access. The same libheif flaw affects platforms like Slack, Meta, GitHub Enterprise, and Zoom. The research cost less than $3,000 in AI tokens and was conducted by three researchers. Only Shopify detected the activity despite thousands of malicious images being sent.
Anthropic Co-founder Jack Clark Says Liberal Arts Degrees Will Thrive in AI Age
Jack Clark, a literature graduate and co-founder of Anthropic, says liberal arts degrees will become more valuable than traditional STEM degrees in the AI era. He believes knowing the right questions to ask across disciplines matters more than rote programming skills. His comments come as AI threatens to displace millions of jobs. Companies like Anthropic and Google's DeepMind are hiring philosophers to handle ethical questions. Critical reasoning and synthesis across subjects are skills AI cannot easily replace.
Huawei Launches Customer-Focused AI Collaboration System for Industries
Huawei introduced a new customer-centric collaboration system called Huawei Plus Partners to scale AI across multiple industries. The system focuses on working closely with customers and industry partners to deploy AI solutions. It aims to help various sectors adopt and implement intelligence at scale. The initiative reflects Huawei's push to expand AI adoption beyond tech into broader business applications.
Former UnitedHealthcare CEO Discusses AI Battle in Health Care
Bill Golden, former CEO of UnitedHealthcare's commercial business, shared insights on AI in health care after 25 years at the company. He now works with Anterior, an AI company focused on health plan administration. Golden warns that AI could make both payers and providers better at fighting each other if underlying systems are not redesigned. He believes AI could create shared data to improve reimbursement and contracting conversations. His campaign Build Faster Horses argues health care must fundamentally rethink processes rather than just optimize them.
CoreWeave Promotes Physical AI for Safety and Innovation
CoreWeave is advancing Physical AI, which uses AI models that interact with hardware in real-world settings. The company acquired Monolith AI to embed engineers with customers in fields like automotive, aerospace, and robotics. Examples include Aston Martin F1 using AI for race strategy and Nissan improving vehicle quality. CoreWeave focuses on extending human capabilities with measurable benefits in safety-critical industries.
Syracuse University Summit Highlights AI and Future Vision
Syracuse University held a summit focusing on AI, interdisciplinary research, and the future of higher education. Chancellor-elect Haynie emphasized the need for change as colleges face declining enrollment and budget pressures. The university showcased an AI bot named Katie and made the chatbot Claude available to over 11,000 campus users. The event also highlighted combining science and humanities to drive innovation.
Debate Grows Over China's AI Distancing From Theft Claims
U.S. companies and officials have accused Chinese AI labs of using distillation to copy advanced models, calling it theft. However, Cohere CEO Aidan Gomez argues China has developed world-class AI capabilities independent of distillation. He notes Chinese models now beat top American models on some benchmarks. Analysts also point to China's large talent pool and government support as drivers of its rapid AI progress.
IMF Warns AI Could Widen Inequality in Europe
An IMF paper presented to EU ministers in Dublin says AI could boost European productivity by about 1% over five years. However, it also warns AI may widen inequality and strain power networks. Around 60% of workers in advanced European economies face high exposure to AI. The IMF recommends deeper EU integration and investment in Europe's own AI industry to reduce reliance on the U.S. and China.
Atlassian Turns Jira Into an AI Coding Command Center
Atlassian rolled out new tools for Jira and its DX platform to help teams monitor AI-driven software development. The tools address concerns about AI agents making unintended changes to codebases. They give engineering teams clear visibility into how AI agents work on code. This move represents a significant step in safely adopting AI in software development.
Sources
- When AI Hacks AI: Inside the Shocking OpenAI Security Breach
- White hat hackers just breached OpenAI using Anthropic's Claude in less than 72 hours — and it is a case study in just how fast AI is advancing
- Anthropic billionaire cofounder Jack Clark studied literature, not code—and says liberal arts degrees will win in the AI age
- Customer-Centric "Huawei + Partners" Collaboration System Launched to Scale AI Across Industries
- Former UnitedHealthcare CEO Addresses AI ‘Battle of the Bots’
- CoreWeave: Physical AI Will Help Us Travel Safely, Not Kill Us All
- Syracuse University’s summit on future: AI, combo classes and other takeaways
- The U.S. says China's AI progress is down to 'distillation.' But is it that clear cut?
- IMF tells EU ministers AI could boost growth but increase economic strains
- Atlassian (TEAM) Just Turned Jira Into an AI Coding Command Center
Comments
Please log in to post a comment.