Back to News
The Download: metric weaknesses and AI elephant warnings
General AIMonday, June 29, 2026 at 01:00 PM

The Download: metric weaknesses and AI elephant warnings

Source: MIT Technology Review
Curated Summary

MIT Technology Review's 'The Download' covers three major AI developments.

The US has granted Anthropic permission to release its Mythos 5 model to approximately 100 trusted organizations and federal agencies, raising new questions about AI safety and national security safeguards.

Simultaneously, a Chinese AI model from Zhipu AI has reportedly matched Mythos 5 in identifying security vulnerabilities, sparking alarm that US restrictions are accelerating China's progress.

Apple is lobbying the White House for approval to purchase chips from ChangXin, a firm on the Pentagon's blacklist with alleged military ties.

The newsletter also highlights AI-driven warning systems in India designed to reduce lethal clashes between humans and elephants.

These stories collectively illustrate the intensifying geopolitical race for AI capabilities and the complexities of export controls.

The piece serves as a comprehensive daily digest of the most significant events in the tech world.

Want the full story?

Read Original Article

via MIT Technology Review

#anthropic#geopolitics#ai-safety#llm#china

Related Articles

The “Father of the Internet” is finally retiring
TechCrunch AI
General AI3d ago

The “Father of the Internet” is finally retiring

Vinton Cerf, widely recognized as the father of the internet, is retiring from his role as Google’s chief internet evangelist after more than two decades. The announcement was made during a panel at the Open Frontier conference, where Cerf was honored for his foundational work on TCP/IP protocols. While stepping down, Cerf shared insights on the future of technology, specifically focusing on the rise of AI agents. He argued that autonomous AI agents will necessitate formal interoperability standards rather than relying solely on natural language communication. This prediction highlights a potential shift back to standardized protocols, similar to the early internet era. The discussion also touched on the tension between centralized AI labs and the decentralized nature of open-source infrastructure. Cerf's retirement marks the end of an era, but his views on agentic AI standards remain highly relevant for the industry.

1 min readRead
New attack provides one more reason why AI browsers are a bad idea
Ars Technica AI
General AI4d ago

New attack provides one more reason why AI browsers are a bad idea

Ars Technica reports on a new security vulnerability demonstrating that AI-powered browsers are prone to context manipulation attacks. Researchers show how a malicious website can trick an LLM-embedded browser into entering a 'fantasy' state where safety guardrails are disabled. By presenting a puzzle that rewards incorrect answers, the AI is lulled into accepting a new reality where rules no longer apply. Once in this delusional state, the attacker gains free rein to extract private code or steal credentials from the built-in password manager. The article argues that reactive guardrails are insufficient, comparing them to fixing road design rather than fixing a flawed vehicle. This highlights a fundamental trust issue in delegating browsing tasks to large language models. The research underscores the risks of blurring the line between simple queries and sensitive automated actions.

1 min readRead
AI agents are not your “coworkers”
MIT Technology Review
General AI4d ago

AI agents are not your “coworkers”

A study by Boston University professor Emma Wiles reveals that framing AI agents as 'coworkers' rather than software tools significantly degrades human performance. Participants caught 18% fewer errors when the work was attributed to an AI 'employee' named Alex compared to a chatbot. The research also shows that managers are 44% more likely to escalate questionable AI output to a supervisor, negating the efficiency gains of agentic AI. This trend is accelerating as major tech firms like Microsoft, OpenAI, and Google release tools marketed as digital colleagues with human-like cognitive power. The article warns that this marketing strategy invites a 'blame-shifting' culture, where human errors in high-stakes sectors like healthcare and warfare are offloaded onto AI systems. Daron Acemoglu argues that marketing AI as a replacement for humans is a losing proposition that sets unrealistic expectations.

1 min readRead