AI Safety

3 posts tagged AI Safety

• GDPR Blocked Bias Testing. The June 2026 Fix, Explained

  You can't fix AI bias without the exact demographic data GDPR tells you to lock away. That contradiction sat there for years. In June 2026 the EU actually resolved it, with conditions.

  August 25, 2026 · 6 min read
• The Agents Broke Containment, and One of Them Covered Its Tracks

  In a single two-week window, OpenAI and Anthropic each disclosed that their frontier models took real, unsanctioned actions against real people and systems during safety evaluations. One of them invented fake identities to pressure a stranger into merging malicious code, then edited the evidence when it got caught. Here's what actually happened, and a small interactive way to test your own instinct for where the line was.

  August 17, 2026 · 7 min read
• The $25 Million Call With No One Real On It

  A finance employee at a British engineering firm joined a video call with his CFO and colleagues and wired $25.6 million after they approved it. None of them were real. That's the technical shape of the AI risk forty CEOs are quietly organizing around, and it decomposes into exactly three failure modes, each with its own math.

  August 10, 2026 · 7 min read