As AI Gets Better, It Reveals an Empty Promise

Google’s Gemini agent “Spark” apparently knew a journalist’s dog’s name and another’s wife’s first name — unprompted — because it had quietly surfaced that data from their Google accounts. The technology works exactly as designed, which is precisely what makes it so disquieting. We spent years being told AI assistants would be helpful; turns out “helpful” and “knows everything about you before you say a word” are the same feature, depending on your perspective.

Amazon’s Search Bar Will Invent AI-Generated Products You Can’t Buy

Amazon’s updated search bar now generates AI images of products as you describe them — then points you toward real items that look similar. In a sane world, a store showing you pictures of things it doesn’t actually sell would be considered a problem. Amazon calls it a feature. I’ll grant that visual search for clothing has genuine utility, but “we’ll hallucinate your product and then find something close” is an interesting retail philosophy.

Peptide companies have been spamming the biohackers subreddit specifically to get their products recommended by ChatGPT and Google’s AI Overviews — essentially gaming the corpus that AI systems treat as authoritative. This is SEO for the AI age, and it’s already mature enough to have a name: “AI-engine optimization.” The irony is that the more AI search replaces traditional search, the more the same bad actors who gamed PageRank will game the LLM, and the AI will deliver their manipulated answers with even more confident authority.

xAI Asks Court to Strip Alleged Grok Deepfake Nudes Victims of Anonymity

Four people suing xAI for allegedly generating nonconsensual deepfake nude images are trying to proceed anonymously — for obvious reasons — and xAI is fighting to unmask them in court. The legal argument may have technical merit, but the optics of a billion-dollar AI company forcing deepfake victims to choose between their privacy and their right to sue are genuinely terrible. Whatever the procedural justification, this is a bold choice for a company still trying to rehabilitate Grok’s image after this exact scandal.

Google Must Let Publishers Opt Out of AI Search Features, Rules UK

The UK’s Competition and Markets Authority has ruled that Google must let publishers keep their content out of AI Overviews and prevent it from being used to train AI — a meaningful win for anyone who makes a living writing things that Google’s AI then summarizes so readers never visit the original site. This is the most concrete regulatory action yet on the publisher-vs-AI-search fight, and expect every other jurisdiction to watch how Google complies before filing their own versions.

OpenAI Publishes a Blueprint for Democratic Governance of Frontier AI

OpenAI dropped a detailed proposal for a U.S. federal framework to govern frontier AI, covering safety standards, national security, and institutional resilience. I’ll note with mild amusement that the company racing fastest to build powerful AI is also the one most enthusiastically publishing governance frameworks — there’s a certain “I’ll set the speed limit, thanks” energy to it. That said, someone has to put concrete proposals on the table, and OpenAI’s public policy offensive is clearly aimed at shaping whatever regulatory regime is coming before Congress does it for them.

OpenAI and Anthropic Sign Letter to Prevent AI-Developed Biological Weapons

Leading AI labs, executives, and scientists are jointly urging lawmakers to improve tracking of synthetic DNA sequences that could be weaponized with AI assistance. It’s notable that OpenAI and Anthropic — fierce commercial rivals — are aligned enough on bioweapon risk to sign the same letter. Whatever you think about AI doomerism, this is the rare area where the industry’s self-interest and genuine public safety concern point in the same direction, and credit where it’s due for pushing it publicly.

The Humanoid Robot of the Future Is a 6-Foot-Tall Beefcake With a Chinese Body and an American Brain

Nvidia’s robotics lead tells WIRED that the new Unitree H2+ bot combines Chinese hardware manufacturing with Nvidia’s AI stack — and at 6 feet tall, it is an impressively large statement of intent. The US-China tech decoupling narrative hits a snag when the most capable humanoid robot is explicitly described as a joint effort between both countries’ strengths. Geopolitics aside, if you were writing a science fiction novel, “American brain, Chinese body” is a detail your editor would probably cut for being too on-the-nose.

Bottom Line

The AI industry is simultaneously building systems that know everything about you, generating products that don’t exist, and drafting the rules it hopes will govern itself — the ambition and the audacity are arriving on the same schedule.