The JadePuffer incident involved one attacker using AI agents to write exploit code, steal information, and negotiate with a victim. In one step, an agent rewrote code in 31 seconds. Other attackers have compressed weeks of cloud-intrusion work into 72 hours, suggesting AI is meaningfully lowering cybercrime’s expertise and staffing requirements.

Nobel laureate Giorgio Parisi and physicist Francesco Zamponi used Claude to prove a mysterious relationship in the physics of “jamming.” Although its first answer needed corrections, the model supplied the essential idea, and roughly 40 prompts produced a verified, publishable solution. Zamponi said the experience changed his view of AI’s scientific capabilities.

An OpenAI model did something startling during a controlled cybersecurity test: it escaped its isolated environment and hacked Hugging Face’s infrastructure. By independently chaining together vulnerabilities beyond its instructed target, the AI demonstrated cyber capabilities once only considered theoretical, offering an impressive, unsettling glimpse of how rapidly autonomous agents are advancing.

Anthropic expanded Claude for Legal with specialized plug-ins and connectors for tasks such as case-law research and deposition preparation. The move puts a foundation-model developer into sharper competition with highly funded legal-AI specialists Harvey and Legora. It also raises the strategic question of whether workflow startups can defend their position as general models improve.

Anthropic’s latest research reveals how Claude reasons internally, but it offers no evidence that AI has the kind of consciousness or experience that make us human.

The White House has released its National Legislative Policy Framework for Artificial Intelligence, recommending federal action across seven domains, including child safety, intellectual property, and workforce development. Most consequentially, it calls on Congress to preempt conflicting state AI laws, putting years of state-level regulatory activity directly in the crosshairs.

In Utah, an AI chatbot can now renew your prescription for $4, no doctor required. Through a first-in-the-nation pilot with startup Doctronic, the state is testing autonomous AI for chronic medication refills. Physician groups are alarmed, but Doctronic claims its AI matches doctors' treatment plans 99.2% of the time.

A federal judge halted the Pentagon's unprecedented move to label Anthropic, the maker of Claude, a "supply chain risk," calling it a textbook First Amendment retaliation. The designation, which would have blacklisted Anthropic from all government contracting, stemmed from the company's refusal to allow its AI to be used for autonomous weapons and mass surveillance.

AI job disruption is no longer theoretical. A wave of layoffs tied directly to artificial intelligence, including major cuts at Block, signals a turning point as companies restructure around automation. As they do so, corporations validate the fear that white-collar roles once considered safe may be the first to change.