> daily_signal(2026_07_18)

San Francisco's city attorney sent Apple and Google cease-and-desist letters this week over 13 AI "nudify" apps — face-swap tools used to generate fabricated nude images of women and girls without consent — and both companies pulled or suspended apps within days.

PickBits Daily Signal · Saturday, July 18, 2026

1. San Francisco's city attorney sent Apple and Google cease-and-desist letters this week over 13 AI "nudify" apps — face-swap tools used to generate fabricated nude images of women and girls without consent — and both companies pulled or suspended apps within days.

City Attorney David Chiu's office identified 13 apps — 8 targeting Apple's App Store, 5 targeting Google Play — built specifically to strip clothing from real photos using AI, part of a nudify-app economy a January Tech Transparency Project report pegged at 705 million downloads and $117 million in revenue. The letters accuse Apple and Google of 'aiding and abetting' the spread of non-consensual intimate images and note both companies have 'likely made millions of dollars in fees' hosting them — an accusation aimed squarely at the platforms' own cut of app-store revenue, not just the developers. Chiu's statement is blunt about the harm: 'These images are used to bully, humiliate, and threaten women and girls,' with victims experiencing severe, documented mental-health consequences. The response speed is the actual story. Apple says it has removed three apps and is terminating those developers' accounts, with four more under review for policy violations. Google says all five Play Store apps named in the letter have been suspended, on top of hundreds of other violating apps the company says it removes routinely. That both platforms could act within days — once a city attorney sent a letter citing a specific California law criminalizing non-consensual deepfake pornography — is itself the indictment: the technical and policy capability to stop this was never the bottleneck. The letters gave both companies 28 days to respond to the city or face civil penalties under state law.

Key fact: IF AN AI-GENERATED IMAGE OF YOU IS CIRCULATING WITHOUT YOUR CONSENT, YOU HAVE MORE LEVERAGE THAN THE APP STORE ITSELF PROVIDES. California's law against non-consensual deepfake pornography is what forced this action — check whether your state has an equivalent (most now do) before assuming there is no legal remedy. Report the specific app to Apple (reportaproblem.apple.com) or Google Play (via the in-store 'flag as inappropriate' + a direct abuse report to Google Play's Trust & Safety) citing non-consensual intimate imagery by name — generic 'inappropriate content' reports move slower than a legally specific complaint. If the images originated from a platform (Instagram, X, Snapchat), most now have a dedicated NCII (non-consensual intimate imagery) reporting flow separate from general abuse reports; use it, and keep dated screenshots as evidence before the content is taken down.

techcrunch.com · tech.yahoo.com · primary source

2. TikTok began testing an opt-in tool this week that scans for AI-generated videos using a creator's face without permission — following creator identity verification through a real-time selfie and ID check — and lets them report impersonating posts, days after YouTube expanded a similar tool to eligible creators.

The tool works in two steps. First, a creator who wants protection verifies their identity through Jumio: a real-time selfie plus an ID check. TikTok says it does not retain the ID documents themselves — the facial data extracted is used only to match the creator's likeness against AI-generated content circulating on the platform. Second, once verified, TikTok scans AI-generated video for matches to that creator's face; when it finds a potential match, the creator reviews it and decides whether to report the specific post or account as impersonation. TikTok US spokesperson Zachary Kizer confirmed the test is currently limited to a small group of US creators — this is a pilot, not a platform-wide rollout. The timing is the tell: TikTok is following YouTube's playbook almost exactly. YouTube built its own likeness-detection tool after months of quiet testing, then expanded it to eligible creators over 18 and extended similar protection to celebrities and talent agencies earlier this year. With the two largest short-video platforms now both fielding likeness-detection tools, this is fast becoming a baseline expectation rather than a differentiator — the open question is whether either platform extends real protection to non-creators, ordinary people whose faces show up in AI-generated content without ever having opted into a verification-and-monitoring system in the first place.

Key fact: IF YOU'RE A US CREATOR CONCERNED ABOUT YOUR LIKENESS BEING USED IN AI-GENERATED VIDEOS, CHECK WHETHER YOU'RE ELIGIBLE FOR THE PILOT NOW RATHER THAN WAITING FOR A WIDE ROLLOUT — small pilots like this often expand by invite before they expand by open enrollment, and being verified early means TikTok is actively scanning on your behalf sooner. Before opting in, understand exactly what the Jumio step captures (a live selfie + ID check) and TikTok's specific retention claim (ID documents not kept, facial data used only for matching) — if that data-handling promise matters to your decision, that is the sentence to hold the company to if the policy ever changes.

digitaltrends.com · primary source

3. The Senate voted 46-50 along party lines on Thursday to block a Democratic-led resolution that would have ended a Medicare pilot letting AI approve or deny care for seniors — keeping the program running in six states with no exit date, even as Sen. Cantwell's own data shows approval times have more than doubled since it started.

The WISeR (Wasteful and Inappropriate Service Reduction) model began January 2026 as a six-year CMS experiment: it brings AI-driven prior authorization — a cost-control tool almost exclusively used in privatized Medicare Advantage plans — into traditional, original Medicare, where the practice has historically been rare. It currently runs in six states: Arizona, New Jersey, Ohio, Oklahoma, Texas and Washington, covering services like skin/tissue substitutes and epidural steroid injections. On Thursday 2026-07-16, the Senate voted on a Congressional Review Act resolution to end it. The vote split exactly on party lines: 50 Republicans blocked the resolution, 46 Democrats voted to kill the program, with four senators — Republicans Bill Hagerty, Mitch McConnell, and Roger Wicker, and Democrat Chris Murphy — not voting. Senators Ron Wyden, Maria Cantwell, Richard Blumenthal and Kirsten Gillibrand led the argument that the model delays care. Cantwell's own analysis of Washington-state data found patients waiting two to four times longer for procedures their doctors recommended — from roughly two weeks before WISeR to four to eight weeks after. CMS Administrator Dr. Mehmet Oz defended the program on the opposite premise: 'CMS is committed to crushing fraud, waste and abuse, and the WISeR Model will help root out waste in Original Medicare.' The White House opposed the Democratic resolution, and the STAT reporting notes many Republicans who otherwise want to rein in Medicare Advantage prior authorization were not willing to back a Democrat-led bill specifically. A separate, earlier track exists in the House: the House Appropriations Committee voted unanimously on 2026-06-10 to attach a rider to HHS's 2027 spending bill blocking funds for WISeR or any similar prior-authorization model in traditional Medicare — but that still needs to clear the full House and Senate before it has any force, and Thursday's Senate vote leaves the program fully operative in the meantime.

Key fact: IF YOU OR A FAMILY MEMBER ARE ON TRADITIONAL MEDICARE IN ARIZONA, NEW JERSEY, OHIO, OKLAHOMA, TEXAS, OR WASHINGTON, AND YOUR DOCTOR RECOMMENDS A SERVICE COVERED BY WISER (SKIN/TISSUE SUBSTITUTES, EPIDURAL STEROID INJECTIONS, AND THE FULL CMS-LISTED SERVICE SET), BUILD IN THE LONGER TIMELINE NOW. Cantwell's data shows 4-8 week waits, not 2 — ask your provider's office to submit prior-authorization requests as early as the recommendation is made, not when the procedure is scheduled. If you're denied, you have a right to appeal; Medicare's standard appeals process (start at Medicare.gov or 1-800-MEDICARE) applies to WISeR denials the same as any other coverage decision, and CMS's own materials on the pilot are required to disclose that an AI system was involved in the recommendation — ask explicitly whether an algorithm, not a person, made the initial call, since that changes what your appeal should target.

statnews.com · newsweek.com · healthcaredive.com · primary source

4. Nobel laureate Jennifer Doudna's lab at UC Berkeley used an AI protein-design model to invent a synthetic CRISPR enzyme that doesn't exist in nature — a gene-editing 'molecular scissors' that matches or beats natural enzymes on editing efficiency and cuts off-target errors, published in Science this week.

Doudna won the 2020 Nobel Prize for co-discovering CRISPR-Cas9; her team at the Innovative Genomics Institute and UC Berkeley just used AI to go further than nature did. Rather than searching for a better-editing enzyme in the wild, the researchers used an inverse protein-folding model — AI that starts from a desired 3D structure and reverse-engineers an amino-acid sequence to produce it — constrained by evolutionary data from the TnpB family of CRISPR-Cas12-like proteins, to design synthetic nucleases from scratch. They call the results SynTnpBs. Cryo-electron microscopy of the best-performing variant showed the AI had introduced new electrostatic and hydrogen bonds at the guide-RNA-to-DNA interface — contacts a natural evolutionary search never found — and in gene-editing assays across animal and plant cells, the top SynTnpB variant matched or exceeded its natural counterpart on editing efficiency while showing the lowest off-target activity of any variant tested. That combination — as good or better at making the intended cut, more precise about not making unintended ones — is the two-variable problem every CRISPR therapy has to solve, and it's the variable off-target editing that has been the main safety brake on turning CRISPR into approved human therapies. The framework itself, not just this one enzyme, is the finding: it establishes a general approach for designing non-natural nucleases, meaning the same AI method can be pointed at the next gene-editing bottleneck rather than this being a one-off result.

Key fact: IF YOU WORK IN GENE-EDITING RESEARCH OR THERAPEUTIC DEVELOPMENT, THE REUSABLE PART OF THIS RESULT IS THE METHOD, NOT JUST THE ENZYME. An inverse-protein-folding-plus-evolutionary-constraint approach that can design a nuclease beating its natural counterpart on BOTH efficiency and off-target rate is a template applicable to other CRISPR-family proteins beyond TnpB — worth evaluating against your own target-specificity bottlenecks, especially if off-target activity has been the blocker standing between your program and a clinical-safety case.

the-scientist.com · primary source