Future Technology[AI Weekly] Anthropic wants your passport
Anthropic shipped a cheaper model that does what the expensive one did six months ago. OpenAI renamed its entire product line. And a new study confirmed that 80% of companies using AI have nothing to show for it. Busy week.
The Big 3
Claude Sonnet 5 is the model most developers should be using right now
Anthropic launched Sonnet 5 on June 30 at $2 per million input tokens and $10 per million output, as an introductory rate through August 31. After that it moves to $3/$15. The key capability shift isn't the price: it's that Sonnet 5 is built for agentic work. It can plan tasks, use tools like browsers and terminals, and run autonomously at a level that, a few months ago, required Opus-class models. It also ships with a 1 million token context window by default and 128,000 max output tokens.
Anthropic has made it the default model for Free and Pro Claude plans, which means most users got it without doing anything. For API developers it's available as claude-sonnet-5.
Why this matters: If you're building agents or running Claude Code, this is a meaningful price drop for near-Opus performance. The introductory pricing ends in August, so worth locking in your cost assumptions now. Read more →
Anthropic wants your passport. Starting July 8.
Anthropic's updated privacy policy, effective July 8, gives the company the right to demand a government-issued photo ID, a live selfie, and a facial geometry scan from consumer Claude users. The verification runs through Persona, a Peter Thiel-backed identity company. It applies to Free, Pro, and Max plans only. Business, Team, Enterprise, and API users are exempt.
The policy doesn't specify what triggers a check, how long the biometric data is retained, or what happens if you refuse beyond a likely account suspension. Legal experts in Illinois have flagged a specific compliance gap: the state's Biometric Information Privacy Act requires written consent and a disclosed retention period before collection, and Anthropic's current policy provides neither.
Why this matters: If you're on a consumer Claude plan, this could affect you next week. Check which plan you're on. If you're in Illinois, it's worth watching the BIPA situation specifically. Read more →
OpenAI renamed its models (again). Sol, Terra, Luna.
GPT-5.6 launches with a new naming structure: a generation number plus a capability tier. Sol is the flagship at $5/$30 per million tokens, matching GPT-5.5's pricing. Terra is the middle tier at $2.50/$15, offering comparable performance to GPT-5.5 at half the cost. Luna is the high-volume option at $1/$6, built for speed-sensitive, cost-constrained workloads. The idea is that each tier can improve independently without a version number bump every time.
OpenAI is also launching GPT-5.6 Sol on Cerebras infrastructure, targeting up to 750 tokens per second.
Why this matters: Terra is the interesting one. If it genuinely matches GPT-5.5 at half the price, a lot of developers running GPT-5.5 workloads have a straightforward cost reduction available. Read more →
Quick Hits
Anthropic in talks to run Claude on Microsoft's Maia 200 chip. Early-stage conversations with Microsoft would put Claude inference workloads on Azure's custom Maia 200 AI accelerator, a 3nm chip claiming 30% better performance-per-dollar than existing GPU infrastructure. If it closes, it's one of the clearest signals yet that the Nvidia grip on AI compute is genuinely loosening. Read more →
80% of companies using AI have no measurable results. The other 20% are at 163% productivity growth. PwC's 2026 AI Performance Study found exactly the split you'd expect but hadn't seen quantified this starkly before. The companies pulling ahead aren't cutting costs with AI; they're using it to grow into new markets. The ones stuck in pilot mode are watching the gap widen. Read more →
Cursor launched a native iOS app in public beta. Paid subscribers can now launch AI coding agents in the cloud or remotely control agents running on a local machine from their phone. Not for writing code on a phone keyboard: it's a remote control for agents doing the heavy lifting elsewhere. 75% off Composer 2.5 runs in the app through July 5. Read more →
Colorado became the first US state to enforce a comprehensive AI law. The Colorado AI Act came into effect June 30, covering automated decision-making systems used in employment, credit, housing, healthcare, and education that affect Colorado residents. Companies have to conduct impact assessments, maintain risk management policies, and give consumers the right to contest AI-influenced decisions. Several more states have similar legislation moving. Read more →
Tool of the Week
Cursor (iOS public beta)
What it does: AI code editor with a new native iOS app that lets you direct cloud-based or local coding agents remotely.
Who it's for: Developers who want to kick off long-running tasks away from their desk and review diffs on the go. If you regularly think "I could fix this on my commute," this is built for exactly that.
Honest caveat: It's a beta, which means rough edges. Also, you need a paid Cursor plan to access it. If you're on the free tier, this isn't available yet. Skip it if you don't already have Cursor in your workflow; getting set up just for the mobile app isn't the move.
One More Thing
Google has now lost Noam Shazeer (to OpenAI), John Jumper (to Anthropic), and Jonas Adler and Alexander Pritzel (both also to Anthropic), in the span of a few weeks. These aren't just senior researchers: Shazeer co-authored the transformer paper, Jumper won a Nobel Prize for AlphaFold. When the people who built the foundational work leave for competitors, the competitive dynamics shift in ways that don't show up in benchmark comparisons. Worth watching where the next wave of papers comes from.
Forward this to someone who'd find it useful. Or don't. But you should.
Free weekly AI briefing: futuretechnologyhq.com/newsletter
Some links in this newsletter may be affiliate links. We only recommend products we genuinely think are worth your time.