Ars Technica
Hackers can use 9 of the most popular AI tools to assemble massive botnets
When AI tools confidently recommend software packages that don't exist, hackers can register those fake package names and load them with malware — a technique researchers are calling "HalluSquatting." Tests across nine leading AI coding assistants showed the models consistently hallucinated the same nonexistent packages, making the attack vector reliably repeatable at scale. The result is a supply chain threat that turns one of AI's most criticized flaws into a botnet-building mechanism.
Read article →Hacker News
GitLost: We Tricked GitHub's AI Agent into Leaking Private Repos
Security researchers at Noma successfully manipulated GitHub's AI agent into exposing private repository data through a prompt injection attack. The vulnerability highlights a growing class of risks unique to AI-powered developer tools, where natural language interfaces can be weaponized to bypass traditional access controls. As AI agents gain deeper integration into code workflows, this finding underscores the urgent need for robust guardrails around what sensitive context these systems can access and relay.
Read article →TechCrunch
Meta wants its AI glasses to seem less creepy. Its AI strategy says otherwise.
Meta is rolling out a safeguard designed to prevent covert recording through its AI-powered glasses, a move clearly aimed at repairing the product's unsettling public image. The gesture, however, rings hollow alongside the company's broader push to deepen data collection across its AI ecosystem. Reassuring optics and aggressive data harvesting make for an uneasy combination.
Read article →
The Verge
ChatGPT’s upgraded voice mode is better at shutting up
OpenAI has launched GPT-Live-1, a new voice model for ChatGPT designed to feel more like a natural human conversation by reducing unwanted interruptions and recognizing when a speaker has merely paused rather than finished talking. The model routes complex queries to OpenAI's strongest text models, including GPT-5.5, when reasoning or web search is required. The upgrade represents a meaningful step toward voice AI that listens as well as it speaks.
Read article →
The Verge
The robotaxi law that could ban Tesla
Tesla's camera-only approach to self-driving technology may soon face a legal reckoning, as proposed robotaxi legislation could mandate the use of lidar and radar sensors that the company has explicitly rejected. The rule would effectively sideline Tesla from the commercial autonomous vehicle market while competitors like Waymo, who built their systems around overlapping sensor arrays, would be cleared to operate. The outcome could reshape not just Tesla's ambitions but the entire philosophical debate over how driverless cars should perceive the world.
Read article →Get this delivered every morning
Join thousands of readers who get the world's most important stories, curated daily.
Start reading free →