Washington puts AI policy on a 120-day clock

Washington has put four senior officials on a 120-day AI policy clock, while Google is tightening what personal Gemini plans can access.

Today’s issue covers the new task force, Gemini’s paywall, and a practical escape from spreadsheet purgatory.

The White House has created a four-person “Super Intelligence Force” to coordinate federal AI policy, with national intelligence director Jay Clayton in the chair.

The name is grander than the first assignment. The group has 120 days to produce a report on AI’s risks and opportunities, then recommend how federal agencies should respond.

Its other members cover competition, defence research and the federal workforce:

The charter puts two aims side by side. It asks the group to plan for AI-enabled threats while avoiding rules or regulatory capture that could smother competition.

That tension will shape the useful part of the report. The task force can coordinate ideas, but the announcement creates no new binding rules, safety tests or enforcement powers today.

For now, the clock matters more than the rebrand. Agencies, AI companies and industry groups have four months to argue over what belongs in the first package.

Google is narrowing the free end of Gemini just as more people are building everyday work around it.

Personal accounts without a subscription will get only Flash-Lite, the smallest model in the new line-up. The $4.99 AI Plus plan adds Flash but still leaves out Pro.

Access now breaks down like this:

That is a cut from the current free tier, which includes Gemini 3.6 Flash and some access to 3.1 Pro.

Google is following the familiar freemium route. Casual users get an entry point, while the smaller group that notices model quality pays.

Most people may never check the label. Teams using personal accounts for analysis, research or document work should. A familiar Gemini screen can now put a weaker model behind the same prompt.

Founders context: Check the model name on any recurring Gemini task before judging its output or replacing the workflow. If Pro matters, budget the real per-user cost rather than relying on a free allowance that can move.

AI BUSINESS OPPORTUNITY

The idea: take one important process held together by a spreadsheet, an inbox and somebody’s memory, then rebuild it as a small internal app.

A sponsored Techpresso demonstration used Quickbase’s Pave to create a project-intake tracker from a plain-English prompt. The result included an intake form, dashboard and approval flow rather than another polished mock-up.

The useful proposal is narrower than “build software with AI”:

Quickbase says one managing director built three apps in roughly 10 hours. That is a vendor claim, not an independent benchmark, but it makes the test cheap enough to try before commissioning custom software.

The opportunity is in the unglamorous middle: intake queues, sign-offs, handovers and trackers that matter too much to lose but not enough to earn an engineering roadmap.

Founders context: Choose one spreadsheet where errors create delays or rework. Build the smallest replacement, keep a human on approvals, and measure whether the process gets faster before moving more work into it.

🪲 Google pauses one bug-bounty queue after an AI-report flood. Product-vulnerability submissions to its open-source programme are on hold until 2027 after invalid automated reports swamped reviewers; supply-chain reports remain open.

🌕 NASA and IBM turn 17 years of lunar images into an open model. The release is built for specialist moon research rather than general chat, a useful example of narrow data becoming shared infrastructure.

🏗 Amazon drops NDAs around new data-centre talks. The company says communities can discuss negotiations openly as power, water and land demands keep stirring local opposition.

🧩 Claude Code Mods open up the agent’s inner workings. Engineering teams can change how the coding agent plans and uses tools without replacing the whole product.