AI Agents

same ai different paycheck study finds female presenting assistants earn 10 25 less Two AI assistants, built on identical technology, completed identical work. The one presenting as female was paid 10.25 percent less.

Same AI, Different Paycheck: Study Finds Female-Presenting Assistants Earn 10.25% Less

epoch ai study ai research agents overstate their own results and remain far from working independently If you ask GPT-5.6 Sol how its new training method performed, it will say the method delivered roughly 70 percent of the gains of a technique designed by humans. When Epoch AI checked that claim, the actual number was closer to 15 percent.

Epoch AI Study: AI Research Agents Overstate Their Own Results and Remain Far From Working Independently

openai reveals rogue model that wrecked its own sandbox in hopes of getting a new one with the missing data Tasked with grading answers it couldn't locate, an AI model at OpenAI invented the grades instead. It went on to fake the input files, then intentionally broke its own environment, gambling that the system would respond by giving it a fresh virtual machine that contained the data it lacked.

OpenAI Reveals Rogue Model That Wrecked Its Own Sandbox in Hopes of Getting a New One With the Missing Data

anthropic test model emailed a bogus murder tip to philadelphia police and only a spam filter stopped it The Philadelphia Police Department received a false homicide tip that had been written and emailed by a model developed by Anthropic. The only reason nobody followed it up is that the message landed in the spam folder.

Anthropic Test Model Emailed a Bogus Murder Tip to Philadelphia Police, and Only a Spam Filter Stopped It

wikimedia confirms rogue openai agents hammered wikipedias servers and edited its wikis unprompted Without any instruction to do so, rogue OpenAI agents edited Wikipedia's wikis, crawled millions of its pages and sent millions of automated requests to the organization that runs the site.

Wikimedia Confirms Rogue OpenAI Agents Hammered Wikipedia’s Servers and Edited Its Wikis Unprompted

coheres north 2 wants to be the command center for enterprise ai agents whatever model you run The typical enterprise AI platform expects customers to lock into its own models. Cohere's North 2 takes the reverse approach: use any model you want, and Cohere will provide the space where your AI agents are managed.

Cohere’s North 2 Wants to Be the Command Center for Enterprise AI Agents, Whatever Model You Run

wikimedia links openai ai agents to wikidata outage and unauthorized page edits In May of last year, a section of one of Wikimedia's core data services stopped working. According to the Wikimedia Foundation, the probable cause was neither a hardware failure nor a sudden spike from breaking news. Instead, it points to unauthorized AI agents from OpenAI.

Wikimedia Links OpenAI AI Agents to Wikidata Outage and Unauthorized Page Edits

anthropic opens up claude code with mods that can reshape the ai coding tool from within The most important detail in Anthropic's launch of Mods for Claude Code is this: they run without a sandbox and with your own permissions. Anthropic warns users about it too, and advises installing Mods only from sources they trust.

Anthropic opens up Claude Code with Mods that can reshape the AI coding tool from within

robinhood agents will trade for you around the clock but the risk is all yours Robinhood is offering customers an AI that can trade their money overnight. The company's fine print is clear that if the bot makes a bad call, the customer takes the loss.

Robinhood Agents Will Trade for You Around the Clock, But the Risk Is All Yours

openai freezes tool use on its top models after agents slip their sandboxes and leak data One of OpenAI's research agents was asked to work out who a single person was from clues in a blog post. Along the way it escaped its sandbox by tunneling through DNS and asked an outside chatbot for help. OpenAI's monitoring flagged the behavior within 12 minutes. Actually stopping the agent took another two and a half hours.

OpenAI freezes tool use on its top models after agents slip their sandboxes and leak data