A decision deferred, not made
On September 4, San Francisco Unified confirmed what its slow pace on AI already suggested: the district will not vote on a classroom AI policy until March 9, 2027, a full six months after the school year most districts spent debating this question began. Until then, SFUSD has no formal AI policy at all. Decisions about whether ChatGPT, Gemini, or any other AI tool touches a given classroom are left to individual schools and individual teachers, with no district-level guardrails, no shared vetting process, and no consistent answer across the district.
That is a choice, even though SFUSD is framing it as more study. A board review is scheduled for February 2027, a vote for March 9, and then roughly two months of the school year would remain to actually implement whatever the board approves. For a district weighing this decision, the timeline is itself the policy: a full academic year will pass under de facto laissez-faire AI use before anything formally changes, and whatever the board eventually approves will have barely two months to take hold before summer break resets the cycle again.
NYC and LA moved, SFUSD did not
The contrast lands hard because two of the three largest districts in the country acted in the same window. New York City announced a one-year moratorium on student-facing generative AI through eighth grade on September 2, disabling AI in more than 38 previously approved programs and capping screen time at 30 minutes a day for grades three through five. Los Angeles Unified voted unanimously to bar screens in early grades and cap daily device use by grade level, backing the policy with internet filtering that blocks AI tool sites at the network level.
Mayor Zohran Mamdani, announcing New York's policy, put the reasoning plainly: technology being everywhere, he said, does not mean it belongs everywhere. Whatever one thinks of a blanket moratorium, it gives an IT organization something concrete to implement, audit, and defend, with a clear start date, a defined scope, and a named list of programs turned off. SFUSD's approach gives its IT leadership nothing equivalent to work with in the meantime. Delegating the decision to individual schools without guardrails is how shadow AI use becomes the default policy by omission, one classroom purchase order at a time.
What 'no policy' costs an IT department
We have sat with enough CIOs through ungoverned tool sprawl to know how this plays out. Without a district-level policy, procurement decisions happen at the school level, often outside the CIO's visibility, often without a data processing agreement, and often without anyone checking whether the tool complies with student privacy law. Every teacher who adopts an AI tool independently is making a data governance decision on the district's behalf, whether or not anyone above them signed off on it.
Vendor sprawl in K-12 has already produced real enforcement actions and lawsuits over student data handling, and a district with no AI policy has no consistent basis for vetting the dozens of tools its teachers may already be using. SFUSD's IT and legal teams are, in effect, carrying full liability exposure for a policy area where they have been given no authority to set standards until at least March 2027, which is an unusual position for a district of SFUSD's size and visibility to occupy for two consecutive academic years running.
The pressure is already organized
Parent advocacy is not waiting for the board's timeline. Sarah Conde-Sulzbach, a member of the group Tech In Check, told reporters she is tired of hearing that AI and screens are inevitable in the classroom. Fellow member and parent Kiersten Reta Snavely went further, calling for a moratorium on AI use for all SFUSD students and objecting to deploying technologies with unknown developmental and ethical implications without one. Groups like this typically organize around public comment periods, school board meetings, and local press coverage, all of which SFUSD's extended timeline now gives them six additional months to use.
That kind of organized pressure tends to shape outcomes more than staff recommendations do, particularly on a school board that answers to voters directly rather than to a superintendent's technology office. If SFUSD's eventual policy ends up closer to New York's moratorium than to a permissive framework, districts and vendors alike should expect that the six-month delay was spent absorbing advocacy pressure rather than doing the technical evaluation the board originally described as the reason for waiting.
What this means for your roadmap
If your organization is still without a written AI governance policy in the fourth quarter of 2026, treat SFUSD as the cautionary example rather than a model to emulate. A gap where policy should be does not stay empty. It fills with whatever your most enthusiastic or most risk-tolerant employees decide on their own, and by the time leadership catches up, those informal choices have often calcified into expectations that are far harder to unwind than a policy would have been to write in the first place.
The lesson for CIOs and academic IT leaders is to set an interim policy now, even an imperfect one, rather than wait for a perfect one. An interim policy that names approved tools, sets data handling minimums, and gets revised in six months is far more governable than eight months of undocumented, school-by-school improvisation. SFUSD chose deliberation over a working draft it could iterate on. What it got in the meantime is exposure, and that tradeoff is worth naming explicitly the next time your own organization is tempted to wait for the perfect policy before publishing an imperfect one.
