Summary: From February 16 to 20, 2026, the roundup connects ChatGPT’s Lockdown Mode, Claude Sonnet 4.6 becoming the free and Pro default, Gemini 3.1 Pro preview’s doubled reasoning at unchanged price, and Meta’s NVIDIA GPU partnership. Together, the items show security entering release cadence while default placement, price-performance, and locked-in compute shape competition.
No single announcement dominated AI news between February 16 and 20, 2026, yet read side by side, four big stories trace one unmistakable direction: security turning into a product feature, the default slot turning into a battlefield, price-performance turning into the shared language, and compute hardening into the moat. Distribution and economics, not raw capability, sit at the center of three of the four — telling evidence of where the contest has moved. A story-by-story walkthrough follows.
ChatGPT Locks Things Down
February 16: OpenAI brought Lockdown Mode and Elevated Risk labels to ChatGPT to counter prompt injection and data-exfiltration attacks. Once an assistant browses pages and touches data on the user’s behalf, everything it ingests from outside becomes a possible attack carrier — folding detection and restriction into the product core, written into the docs and versioned alongside releases, is the road any product of this scale eventually has to take. Labels paired with lockdown add up to defense-in-depth, delivered as a shipping feature.
Claude Sonnet 4.6 Moves Into the Default Seat
February 17: Anthropic shipped Claude Sonnet 4.6 and slid it straight into the default-model seat for free and Pro users alike, with knowledge cutting off in May 2025. What the default seat means: the majority of users never pick a model — the product picks for them — which makes it the priciest square on the board at every generational handover, and the most fiercely fought. Free-tier users running the newest model from day one leave no doubt about the acquisition play at work.
Gemini 3.1 Pro Talks Through Its Price Tag
February 19: Google put Gemini 3.1 Pro into preview with 2x the reasoning of Gemini 3 Pro at an unchanged price, on the gemini-3.1-pro-preview API endpoint. With capability gaps between labs getting harder to tell apart, Google shifted the axis of comparison to capability per dollar — the straightest route from infrastructure scale to monetization. For API teams, that arithmetic lands directly in unit economics.
Meta and NVIDIA Ink a Million-GPU Contract
Also this week, Meta took to its official newsroom to announce a long-term infrastructure partnership with NVIDIA covering millions of Blackwell GPUs. Locking up compute now works as a moat in frontier competition — a multi-year commitment counted in millions pushes everyone else’s cost of entry up one more rung. Capacity plans that still assume renting at scale on short notice are now positioned against the direction every hyperscaler is locking in.
The Week at a Glance
- Security joined the release cadence: with Lockdown Mode and Sonnet 4.6 landing in the same week, defense stops living in a docs appendix and starts shipping inside the version
- Default placement and price worked as the week’s twin levers: Anthropic claimed the default slot; Google fixed the price-performance tone with a doubling that kept the price flat
- Infrastructure’s unit of scale climbed once more: from model races to GPU contracts, the cost floor of competing keeps rising
- Platform defaults set the tempo: one switch hands an entire user base a new model, so vendor confidence — and vendor risk — scale right along with it
- Second-order effects deserve watching: risk labels reset what enterprise buyers expect from every assistant vendor, and preview pricing reshapes every rival’s pricing page
- Three notes for developers: put preview endpoints through their paces before committing, run regression tests whenever a default model changes, and add injection defense to your agent requirements list