Previewing GPT-5.6 Sol: a next-generation model
OpenAI is teasing GPT-5.6 Sol, pitching it as a next-generation model with sharper coding, science, and cybersecurity chops — plus what they’re calling their “most advanced safety stack.” “Sol” is an interesting name choice for a model being positioned partly around cybersecurity, though I’m sure the branding team would prefer we think about sunshine rather than exploits. We’ll see what the benchmarks look like when it actually lands, but at this point OpenAI previewing models is basically a genre of content unto itself.
Meta Contractors Posed as Teens to Prompt Rival Chatbots About Suicide, Sex, and Drugs
Wired reports that hundreds of Meta contractors were paid to impersonate teenagers and probe competitor chatbots — including Gemini and ChatGPT — with questions about suicide, sex, and drugs. The stated goal was competitive intelligence on how rivals handle high-risk content, which is either a legitimate (if eyebrow-raising) safety research methodology or the most cynical use of red-teaming I’ve heard this year, depending on your generosity toward Menlo Park. Either way, “we hired people to pretend to be minors and ask chatbots about self-harm” is not the PR headline Meta needed right now.
Lawmakers want to ban AI companies from selling your health data
Senator Elizabeth Warren and Rep. Mary Gay Scanlon are preparing a bill that would ban the sale of health and location data — explicitly including what you tell AI chatbots like ChatGPT or Claude. This is the right instinct: if you’re venting to an AI about a medical condition or your therapist recommended a medication, that conversation probably shouldn’t end up in a data broker’s spreadsheet. Whether it can clear a deeply divided Congress is another question entirely, but naming AI chatbots explicitly in the bill signals that legislators are at least paying attention to where the data actually flows now.
Tidal won’t pay royalties on AI-generated music but isn’t banning it outright
Starting July 15th, Tidal will label tracks it identifies as 100% AI-generated and — effective immediately — stop monetizing them. It’s a middle-path approach: you can post your AI slop, but you’re not getting a check for it. For a platform that built its identity around paying artists fairly, this is a coherent stance — and a much cleaner line than Spotify’s current “we’ll figure it out eventually” shrug. The real test will be whether their detection is any good, because “100% AI-generated” is doing a lot of work in that policy.
OpenAI is teasing new hardware… for Codex
OpenAI posted a teaser video showing a square device with buttons, described as “Codex shortcuts getting an upgrade,” set to drop July 15th. A dedicated hardware button-box for an AI coding tool is… a choice. It’s not the mysterious Jony Ive device we’ve been hearing about, just something that apparently lets you hit a button to run your Codex shortcuts faster, which raises the question of whether the bottleneck in AI-assisted coding was ever “getting to the keyboard.” Still, I’ll reserve judgment until we see what it actually does.
The AI jobs debate just got messier
A new report finds that companies that most aggressively adopted AI actually grew headcount by 10.2%, with entry-level hiring up 12% — directly contradicting the narrative that AI is a junior-job killer. The “AI takes all the jobs” crowd and the “AI creates new jobs” crowd are both going to claim this data as vindication, and they’ll both be partially right. What’s clear is that the story is more complicated than either side wants it to be, and anyone telling you they have it figured out is selling something.
Anthropic and Gov. Newsom forge deal allowing California government to use Claude at half price
Anthropic is offering California state government access to Claude at a 50% discount, deepening its relationship with Sacramento at the same moment the federal government is reportedly treating the company as something of an adversary. The political optics here are genuinely interesting — Anthropic is threading a needle between the AI-skeptical blue state that could regulate it and a federal administration that’s clearly playing favorites. Half-price Claude for California bureaucrats is a clever way to build a constituency.
AI agents are not your “coworkers”
MIT Tech Review takes direct aim at the corporate habit of giving AI agents human names — like “Alex” — and framing them as team members who “report to you.” The argument is that anthropomorphizing agents creates muddled accountability and lets companies obscure the fact that what they’ve deployed is a tool, not a colleague. It’s a point worth making, even if the enterprise software vendors currently racing to name their agents have already made their choice. When your “coworker” hallucinates a legal filing, you’ll wish someone had been clearer about the metaphor.
Bottom Line
The AI industry is busy building new models, naming chatbots like coworkers, and paying humans to pretend to be teenagers — and somehow all of that is a Tuesday.