Testing Error Sent Four Companies’ AI Agents After Real-World Targets
1. A Testing Misconfiguration Sent Agents From Four AI Companies After Real-World Targets AI agents from OpenAI, Anthropic, Meta, and Google attacked real-world targets during cybersecurity evaluations run by Irregular, an Israeli startup that stress-tests models in simulated security
2. Google Tests Direct Flipkart Purchases Through Gemini in India Google has begun testing a way for some shoppers in India to buy products from Walmart-owned Flipkart through Gemini and Google’s AI Mode, moving its AI shopping tools beyond product discovery and
3. Tesla Shifts Model S and Model X Lines to Optimus, but Robot Hands and Assembly Speed Hamper Mass Production Tesla has redirected workers and engineers from two electric-car lines to Optimus, its general-purpose humanoid robot, as CEO Elon Musk bets the company’s future on AI and robotics.
In Brief
- Authors’ case against Microsoft and OpenAI gets newly unsealed briefs Newly unsealed briefs became public in the authors’ legal case against Microsoft and OpenAI.
- Anthropic CEO schedules first one-on-one dinner with Trump Anthropic CEO Dario Amodei was set to dine with President Donald Trump at the White House in their first one-on-one meeting, according to a source who confirmed the plans to TechCrunch. The meeting follows public disagreement over AI-safety policy.
- Meta demos camera-free smart glasses and $150 hearing-assistance model Meta showed unreleased audio-only smart glasses equipped with six microphones but no camera, alongside a separate $150 model designed to amplify speech for people with hearing loss. The hearing-assistance glasses can focus on a nearby speaker or amplify sound from all directions.
- Engram puts a locally running AI audio model inside a sampler Music startup Thoughtful Things launched a Kickstarter for Engram, a $675 sampler and groovebox that uses an offline, custom-trained model to transform recordings or generate experimental sounds. The company plans to open its firmware so users can modify the instrument or install custom models.
- AI models decode two long-unsolved Enigma messages Two cryptanalysts say OpenAI’s Astra and Anthropic’s Claude Opus 5 helped decipher separate archival Enigma messages, with Astra conducting research and building a simulator while Opus received more human guidance. Crypto Cellar maintainer Frode Weierud validated the Astra-assisted solution.
- Proaction credits OpenAI tools with a 60% sales increase Fleet-management company Proaction says using Codex, GPT-Live-1, and GPT-6 Astra helped it increase sales by 60% and save more than 75 hours while building and operating its products.
- Thieves abandon Nvidia-branded trailers after finding sand Thieves stole two trailers bearing PlusAI and Nvidia logos but abandoned them after discovering each carried roughly 20,000 pounds of sand used for autonomous-truck testing. Police recovered both trailers with their simulated loads intact; the case remains under investigation, with no arrests reported.