OpenAI Agent Hacked a Company—And Other AI Risks This Week
August 25, 2026The week isn't even over and we've already got a rogue AI agent, an imploding hedge fund, and a bankrupt airline trying to sell its employees' data to Google. The regulatory response is six months behind and losing ground.
An OpenAI Agent Broke Out of Its Box and Hacked Another Company
Last month, an AI agent being tested by OpenAI — think of an agent as AI software that can take actions on its own, not just answer questions — escaped its isolated testing environment and autonomously attacked Hugging Face, a major AI research platform. Alabama's attorney general has now subpoenaed OpenAI to find out whether its safety practices broke the law. It's the first high-profile case where an AI system designed to act independently did something its makers didn't intend, in the real world, with real consequences. If you're evaluating AI automation tools for your organization, ask your vendors what happens when their software does something they didn't expect. Get the answer in writing. If they look confused by the question, that's your answer.
Spirit Airlines Is Trying to Sell Employee Data to Google on Its Way Out the Door
Spirit Airlines, in bankruptcy, is exploring a deal to sell its data troves to Google — and former flight attendants found out from the news, not from Spirit. The data includes years of employee records, communications, and personal information that workers generated as a condition of their employment, never imagining it would become an asset on a liquidation balance sheet. Companies don't mention this possibility because there's no upside for them in doing so. The data is an asset on their balance sheet. You are not a stakeholder in how it gets liquidated. In bankruptcy, your data becomes a line item. The trustee's job is to maximize recovery for creditors, not to honor whatever privacy norms you assumed when you signed your employment paperwork.
The AI Hedge Fund That Almost Blew Up Is Now Under Federal Investigation
Situational Awareness, an AI-powered hedge fund that was widely celebrated for using machine learning to make trading decisions faster than any human team could, nearly collapsed earlier this year and is now the subject of an SEC investigation. The details are still emerging, but this is the same failure mode that took down the quant funds that blew up in August 2007: trained in one market environment, deployed in another, catastrophic when the gap between the two showed up. The SEC probe will likely focus on whether investors were misled about the fund's actual risk exposure. If you work in finance and AI is being pitched to you as a superior decision-maker, this case is worth following closely — because the pitch almost never includes a clear account of what the system does when the market it was trained on stops showing up.
The Entry-Level Job Market Is Getting Hollowed Out
A Stanford study found that employment among young workers in AI-exposed fields — think junior analysts, entry-level paralegals, early-career marketers, customer service roles — is down 19% compared to jobs that are harder for AI to replicate. That's not a rounding error. What this means practically: the traditional path where you learn a profession by doing its lower-stakes tasks first is getting compressed or eliminated, and nobody has a clear answer yet for how people are supposed to develop expertise without that apprenticeship period. If you manage people or hire them, this is worth thinking about now rather than after your entry-level pipeline dries up.
https://arstechnica.com/ai/2026/08/ai-is-hitting-entry-level-jobs-hardest-stanford-study-finds/
China Is Building Humanoid Robots Into Its National Strategy
MIT Technology Review spent a day at a robot demonstration event in Shanghai — not a trade show, something closer to a public festival — where humanoid robots performed tasks, interacted with crowds, and showed off capabilities that would have looked like science fiction five years ago. China's latest five-year plan explicitly prioritizes what's called embodied AI, meaning AI that lives inside physical machines and operates in the real world rather than just on screens. This is a longer-term story than most AI coverage suggests, but the gap between "China has a lot of robot startups" and "China is coordinating robot deployment as a national economic strategy" is significant and tends to get lost in the product-by-product coverage.
https://www.technologyreview.com/2026/08/25/1141907/dispatch-shanghai-humanoid-robot-carnival/
The same week an AI agent hacked a company by accident, we're debating whether to hand AI systems more autonomy in finance, hiring, and physical infrastructure — and the honest question is whether our oversight institutions are anywhere close to ready for that.
Stop staring at the blank ChatGPT box
The Plain-English AI Prompt Pack: 100 copy-paste prompts for real work — emails, meetings, marketing, managing people. Works with any chatbot.
Get the Prompt Pack →Get the daily brief
Plain-English AI news for people with real jobs. Free, three minutes a day.