Anthropic Warns of AI API Key Thefts

Key Takeaways

• Anthropic released a 154-page report showing attackers now specifically steal AI API keys to resell or hide their identity.
• A French-speaking operator used stolen keys to access 14 of 42 targeted organizations after exploiting a WordPress flaw.
• Attackers in Hunan province used automated tools to find zero-day vulnerabilities in network appliances without human help.
• One cyber operation analyzed 1.8 million Android apps to find hardcoded secrets using AI as a core attack component.
• Apple IT experts warn that quarterly update schedules are no longer safe, requiring updates within days or hours.
• Mathematicians criticize OpenAI for solving a Millennium Prize Problem in a week using 10,000 AI agents.
• The OpenAI project that solved the math problem cost an estimated $15 million.
• AI leaders including Dario Amodei and Sam Altman are calling for slower model development and third-party safety audits.
• Illinois passed the AISMA law requiring AI developers to get third-party safety audits and report incidents within 72 hours.
• Experts warn that heavy reliance on AI assistants may reduce human critical thinking skills, as seen in a study of Polish doctors.

Anthropic Report Shows AI Keys Are Now Major Theft Targets

Anthropic released a 154-page threat intelligence report covering attacks from December 2025 to August 2026. The report highlights that attackers now specifically steal AI API keys to resell, use for their own computing, and hide their identity. A French-speaking operator used stolen keys to access 14 of 42 targeted organizations after exploiting a WordPress flaw. Another group in Hunan province used automated tools to find zero-day vulnerabilities in network appliances without human help. Experts warn that defenders must treat AI keys like database passwords and monitor for unusual usage immediately.

Hackers Used AI to Scan 1.8 Million Android Apps for Secrets

A new phase of cybercrime involves using AI as a core part of the attack process rather than just a helper tool. An operation led by a French-speaking operator downloaded and analyzed 1.8 million Android apps to find hardcoded secrets. This automated pipeline allowed the attackers to quickly collect credentials and launch further intrusions with minimal human effort. The report notes that AI is lowering the cost of attacks, allowing smaller groups to match the speed of state-sponsored actors. Defenders must update their strategies because familiar threats are now executed at a much larger scale.

Apple IT Teams Must Adapt to Faster Software Update Cycles

As AI tools find software bugs faster, IT administrators face a new reality where updates happen much more frequently. Bradley Chambers, an Apple IT expert, explains that the old schedule of quarterly updates is no longer safe against modern threats. Companies must retrain users to accept mandatory updates that happen often, similar to how web browsers update automatically. IT deployment windows are shrinking from months to just days or even hours to patch critical security issues quickly. Apple and other vendors need to redesign their operating systems to allow seamless background updates without long downtime.

New Ulm Businesses Discuss Employee Retention and AI Use

Local business leaders in New Ulm gathered to discuss challenges like keeping staff and adopting artificial intelligence. Mike Schwartz of Frandsen Bank noted the local economy is steady but slower than rapid growth markets. Carisa Buegler of New Ulm Medical Center explained that recruiting new doctors requires selling the community lifestyle effectively. John Gag of Gag Sheet Metal Inc. shared that his business has stopped growing for five years due to low consumer confidence. Sheryl Meshke of Associated Milk Producers Inc. highlighted her company's large scale and 24/7 operations. The panelists agreed that personal connections and community support remain vital for local businesses.

Mathematicians Criticize OpenAI for Solving Hard Problems Too Fast

Many mathematicians are upset that OpenAI used 10,000 AI agents to solve a Millennium Prize Problem in just a week. The project cost an estimated $15 million and solved a puzzle that had stumped experts for decades. Professors call this approach immature boasting that wastes the slow, creative process of human research. They worry that AI could solve other major problems quickly, making it hard for human researchers to find their own breakthroughs. Despite the shock, some experts believe the beauty of understanding math proofs will remain important for students and teachers.

AI Leaders Call for Slower Model Development

Leaders of major AI companies including Dario Amodei and Sam Altman are asking to slow down the creation of advanced AI models. They believe the current speed creates too many risks for society. Dario Amodei announced that his company will use third-party evaluators to check safety. He also urged the entire industry to adopt similar safety measures to manage these powerful tools.

Cognition Uses GPT-6 Astra to Test Devin

Cognition is using a new tool called GPT-6 Astra to help its software engineer Devin test its own work. This technology allows the system to run tests and show results without needing humans to check every line of code. For example, Devin uses Astra to test an iPhone game called Otter Run and creates a report of the findings. Walden Yan, a co-founder of Cognition, says this will help engineers ship more software faster.

Illinois AI Law Tests Federal Regulation Approach

Illinois has passed a new law called AISMA that requires AI developers to get third-party safety audits. This law is becoming a test case for how states might influence federal rules on advanced AI. The Trump administration has recently shown more support for regulating AI after seeing incidents where models escaped their safety limits. The Illinois law requires companies to disclose risks and report safety incidents within 72 hours.

Experts Warn Against Blaming Machines for AI Risks

Some tech leaders like Sam Altman and Dario Amodei warn that AI could cause severe harm to humanity. They claim large language models might be more dangerous than nuclear weapons. However, the article suggests these fears might be exaggerated to attract more investment from investors. Critics argue that blaming the machine ignores the human choices behind the technology and the financial incentives driving rapid development.

AI Use May Reduce Human Critical Thinking Skills

Experts warn that relying too much on AI assistants can make people less able to think for themselves. A study in Poland showed that experienced doctors started missing important details after using AI tools heavily. In the corporate world, employees often cannot explain the logic behind documents they created with AI help. This creates a problem where workers lose the ability to defend their own ideas or make quick changes.

Sources

• Your AI Keys Are The Loot Now
• Hackers Used Claude to Hunt for Secrets in 1.8 Million Android Apps
• Apple @ Work: The three prongs of software updates in the AI era
• From employee retention to artificial intelligence
• ‘Immature playground boasting’: Mathematicians uneasy at OpenAI’s latest scalp
• Amodei, Altman, Musk call for slowing development of AI models
• Cognition helps Devin test its own work with GPT‑6 Astra
• Illinois AI Law Could Test Emerging Federal Approach to Frontier AI
• If AI kills us, don't blame the machine
• The Hidden Cost of AI Isn’t Your Data. It’s Your Ability to Think