🏛️ How big tech got its way on Trump’s AI executive order ↗
Trump pulled back from a planned AI executive order that would have pushed safety reviews for new models. The reason, in essence: don’t slow America down in the race with China.
The notable part is that the order was reportedly fairly mild already - voluntary, not some sweeping regulatory hammer. Still, tech leaders pushed hard against it, and they got what they wanted. (The Guardian)
🛡️ Claude Mythos AI Finds 10,000 High-Severity Vulnerabilities in Widely Used Software ↗
Anthropic said its Project Glasswing effort found more than 10,000 high- or critical-severity vulnerabilities using Claude Mythos Preview. That’s impressive, but it also feels like a fire alarm wearing a cape.
The company says 1,726 findings have been confirmed as true positives, with 97 patched upstream and 88 advisories issued. The hard part now is not finding bugs - it’s fixing them before attackers get the same toys. (The Hacker News)
🛰️ White House approves $9B for spy agencies to catch up on AI ↗
The White House approved a secret $9 billion request to buy advanced chips for US intelligence agencies, aimed at running frontier AI models for classified work.
Congress still needs to approve the larger funding, but $800 million is being reprogrammed for faster compute access. The subtext is hard to miss: even spy agencies are now stuck in the AI chip queue, which is both absurd and predictable. (Inquirer.com)
🪞 ‘You can’t control everything’: the rise in plastic surgeons asked to create ‘AI face’ ↗
Plastic surgeons say more patients are bringing in AI-edited selfies as surgery goals - sharper cheekbones, perfect symmetry, poreless skin, the whole uncanny shopping list.
Surgeons warn that AI can rearrange pixels, not anatomy. One reviewed AI-recommended procedures that could run past £100,000 and still not deliver the fantasy face. Grim little mirror maze, this one. (The Guardian)
💰 Anthropic set to close over $30 billion round as soon as next week ↗
Anthropic is reportedly close to a funding round above $30 billion, at a valuation north of $900 billion. Yes, that number is doing cartwheels.
The round is being driven by huge demand for Claude across coding, business workflows and cybersecurity. The company is also racing to lock down more compute, because in AI now, money is mostly fuel wearing a suit. (Business Standard)
🖼️ ChatGPT Images Carry Invisible AI Markers Anyone Can Detect: What Users Who Can’t Disclose Gen AI Need to Know ↗
OpenAI image outputs now carry two provenance signals: C2PA metadata and Google DeepMind’s SynthID watermark. One gives richer file context, while the other is meant to survive screenshots and compression.
There’s also a public verification tool in preview. Not foolproof, no - researchers have shown watermark attacks - but it makes “I totally didn’t use AI” a far riskier little fib. (Tech Times)
FAQ
Why did Trump pull back from the planned AI executive order?
The article says the planned AI executive order was reportedly softened after pressure from big tech leaders. The order was already described as relatively mild and voluntary, with a focus on safety reviews for new AI models. The central argument against it was that added review processes could slow the US in its AI race with China.
What did Anthropic’s Claude Mythos find in software?
Anthropic said its Project Glasswing effort used Claude Mythos Preview to identify more than 10,000 high- or critical-severity vulnerabilities in widely used software. The article notes that 1,726 findings had been confirmed as true positives, with some patches and advisories already issued. The broader concern is whether defenders can fix these weaknesses before attackers exploit similar AI tools.
Why are US spy agencies asking for more AI chips?
The article says the White House approved a secret $9 billion request to help US intelligence agencies buy advanced chips for frontier AI models. The goal is to support classified AI work that requires large amounts of compute. It also notes that Congress still needs to approve broader funding, while some money is being reprogrammed for faster access.
What does “AI face” mean in cosmetic surgery?
“AI face” refers to patients bringing AI-edited selfies to plastic surgeons as examples of how they want to look. These images may show sharper features, perfect symmetry, or skin that is not realistic. Surgeons quoted in the article warn that AI can change pixels with ease, but real anatomy has limits that surgery cannot always overcome.
How do invisible markers on ChatGPT images affect users?
The article says ChatGPT image outputs now include provenance signals such as C2PA metadata and SynthID watermarking. These markers can help identify AI-generated images, including in cases where someone may not want to disclose AI use. The article also notes that watermarking is not foolproof, but it makes undisclosed AI image use easier to detect.