The police report about you may have been written by AI first — and in 48 states, nobody has to tell you. A police report is the official version of what happened — what your insurer reads after somebody backs into your car, what your witness statement becomes once it's sworn. Increasingly the first draft isn't written by the officer — it's written by software that listened to the body camera audio. Here's what these tools do, which two states force disclosure, and how to check your case. The main tool is Axon's Draft One — Axon makes the Taser and most American body cameras. It turns body-camera audio into a transcript and writes a narrative from it — audio only, not video. An officer then reviews, corrects and signs it. It's assistive, not autonomous. Axon says about 600 departments use it across about 600,000 reports, per Forbes in July 2026 — though those are Axon's own figures. The problem is transparency. In July 2025 the Electronic Frontier Foundation filed records requests and found the original AI draft — the version before any officer touched it — wasn't retained. An Axon product manager said so on video: "we don't store the original draft, and that's by design." If that draft is gone, nobody can tell which sentences came from a machine and which a human actually stood behind. Not you, not your insurer, not a judge. Then the laws arrived. Utah went first in 2025: if generative AI helped write a report, it must say so, and the officer must certify they reviewed it. California's SB 524, in force since January 1, 2026, goes further — the report must name the specific AI program and carry the line "This report was written either fully or in part using artificial intelligence," the agency must keep that first draft as long as the report, and an audit trail must record who ran the tool. Two states. That's the whole list. Connecticut's Chief State's Attorney separately paused statewide use in April 2026 — a pause, not a ban. Axon's position: accountability stays with the officer — every report is reviewed and approved by a human, and by default the tool is limited to lower-level incidents. Customers cite big time savings, including a Colorado sergeant reporting an 82% drop. The counterweight: the only randomized trial on this — 85 officers, in the Journal of Experimental Criminology — found AI assistance did not significantly change how long reports took. WHAT THIS MEANS FOR YOU: contact the records division of the agency that took the report — city police, county sheriff or state patrol — with the case number and date. Expect a fee, redactions, and delay if the case is open. Check the footer: Utah and California require a disclosure line, and California's should name the software. If you're charged, that's your attorney's job — but in California that draft and audit trail now exist as records your lawyer can request. One caveat: Axon added draft retention, but it's a SETTING each department switches on. This isn't a story about AI framing anyone — nobody has shown that. It's about auditability: whether anyone can check the official record afterward. Should every state require that first draft be preserved, or is an officer's signature enough? Tell me in the comments. If you want AI news translated into plain English, like and subscribe. Patreon link below to support the channel directly. Keywords: AI police reports, Axon Draft One, California SB 524, Utah SB 180, body camera AI, AI written police report, police report disclosure law, AI transparency policing, how to get a police report, AI in law enforcement, police accountability technology, AI audit trail 👇 Connect with The AI Guide: Website: www.theaiguide.ai YouTube: https://www.youtube.com/channel/UCamFJyTjb_kqUbNpVUVwQeg Patreon: www.patreon.com/theaiguide Facebook: www.facebook.com/davidtheaiguide Instagram: www.instagram.com/theaiguide LinkedIn: https://www.linkedin.com/company/the-ai-guide-on-youtube/ 🔗 Sources: [1] Tech Policy Press (Laperruque, Aug 2026): https://www.techpolicy.press/will-ai-sycophancy-contaminate-law-enforcement/ [2] California Penal Code § 13663 (SB 524): https://california.public.law/codes/ca_penal_code_section_13663 [3] Utah S.B. 180 (2025): https://le.utah.gov/~2025/bills/sbillenr/SB0180.pdf [4] EFF on Draft One transparency: https://www.eff.org/deeplinks/2025/07/axons-draft-one-designed-defy-transparency [5] Axon on AI-assisted reports: https://www.axon.com/blog/how-draft-one-upholds-transparency-for-ai-assisted-police-reports [6] Adams et al., J. Experimental Criminology: https://link.springer.com/article/10.1007/s11292-024-09644-7 [7] Forbes (Brewster, July 2026): https://www.forbes.com/sites/thomasbrewster/2026/07/22/axon-says-ai-police-reports-save-time-public-records-show-they-get-facts-wrong/ [8] GovTech, Connecticut pause: https://www.govtech.com/artificial-intelligence/connecticut-pauses-ai-use-to-create-criminal-reports #ArtificialIntelligence #AIPoliceReports #DraftOne #AITransparency #TheAIGuide

About 1,337 people who BUILD frontier AI just asked the U.S. government to help them slow it down. Their own bosses signed it. This one is different from every other open letter about AI. It didn't come from outside critics or worried academics. It came from inside the labs. More than a thousand employees at OpenAI, Anthropic, Google DeepMind, Meta, Thinking Machines and Safe Superintelligence signed a public statement called "Pacing the Frontier," and the count has kept climbing since it went live in late July 2026. Import AI, Jack Clark's newsletter, covered it in issue #467 on August 3, 2026 under a line that lands hard: after the warning shots come the pleas. The names on it are the point. Dario Amodei, Anthropic's CEO. Jakub Pachocki, OpenAI's chief scientist. Mark Chen, OpenAI's chief research officer. Jared Kaplan and Chris Olah, Anthropic co-founders. Shengjia Zhao, Meta's chief AI scientist. Anca Dragan, Google's VP of AI safety. Shane Legg, a DeepMind co-founder. These aren't junior staffers venting. These are the people paid to make the technology go faster, publicly asking for a mechanism to make it go slower. The ask itself is one sentence. They want the U.S. government to support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development. Read that phrase carefully: automated AI development. They are not talking about chatbots writing emails. They are talking about AI systems that improve themselves, running the research loop that builds the next model, faster than any human review process can follow. And that is not hypothetical anymore. Anthropic has said that as of May 2026, Claude authored more than 80% of the code merged into its own production codebase, up from low single digits before early 2025. The tool is now writing the thing that builds the tool. Once that loop tightens, the honest question stops being what AI can do and becomes how fast it changes between the version you tested and the version you shipped. Here's the part worth sitting with. The statement does NOT ask for a pause, and it doesn't ask any company to stop shipping. Its logic is about competition: no lab can slow down on its own, because if one steps off the gas, the others keep going and take the lead. Unilateral restraint loses. So they want an outside referee with real verification tools, something enforceable rather than a promise. OpenAI and Anthropic endorsed it at the company level within a day. Not everyone is convinced. Mark Zuckerberg published a Wall Street Journal op-ed arguing the opposite, that spreading AI capability widely is safer than concentrating it. Critics also note the obvious tension: the same labs urging caution are shipping the capabilities they say need pacing, and government-built pacing tools could lock in the current leaders. WHAT THIS MEANS FOR YOU: nothing changes in your apps tomorrow. What changes is the signal. When the engineers, chief scientists and CEOs building a technology publicly say they cannot govern its pace on their own, that's a real data point. Practically, expect AI policy to become a live election issue, expect verification and audit requirements in enterprise contracts, and expect the tools you use at work to change faster than your training budget can keep up. If your job touches software, documents or analysis, the gap between the version you learned and the version deployed keeps shrinking. So tell me in the comments: do you trust the people building AI to help decide how fast it moves, or does that job belong to somebody else entirely? I read every reply. Like, subscribe, and hit the bell to stay ahead on AI. Share this with someone who'd want to see it. Support the channel on Patreon (link below). Keywords: artificial intelligence, AI news, Pacing the Frontier, OpenAI, Anthropic, Google DeepMind, Dario Amodei, AI regulation, AI safety, AI policy, AI explained, AI for beginners, recursive self improvement, AI slowdown, tech news 👇 Connect with The AI Guide: Website: www.theaiguide.ai YouTube: https://www.youtube.com/channel/UCamFJyTjb_kqUbNpVUVwQeg Patreon: www.patreon.com/theaiguide Facebook: www.facebook.com/davidtheaiguide Instagram: www.instagram.com/theaiguide LinkedIn: https://www.linkedin.com/company/the-ai-guide-on-youtube/ 🔗 Sources: [1] Import AI 467 — Jack Clark, Aug 3, 2026: https://importai.substack.com/p/import-ai-467-self-sustaining-ai [2] Pacing the Frontier — official statement: https://www.pacingthefrontier.com/ [3] More than 1,200 AI workers ask Washington for help — Fortune, Jul 29, 2026: https://fortune.com/2026/07/29/anthropic-deepmind-openai-meta-washington-ai-slowdown-plan/ [4] OpenAI, Anthropic Formally Back Plan to Slow AI — TechTimes: https://www.techtimes.com/articles/322125/20260729/openai-anthropic-formally-back-plan-slow-ai-that-writes-its-own-code.htm #ArtificialIntelligence #AINews #OpenAI #Anthropic #TheAIGuide

Somebody rented out their face last month for about $74 — not a photo of it, the face. A market for human likeness now exists, with price tags. Platforms in China are paying ordinary people $15 to $700 to license their likeness for AI-generated video. Rest of World reports how the catalogs work: you upload photos or sit for a studio shoot, your face enters a browsable catalog, and producers shop by gender, age and category — labels like "girl-next-door," "rugged" and "supermodel." You pick where it may appear and set your price. Sources: Rest of World, Chinese court records, Deadline, Senate Judiciary. The numbers are small and specific, which is what makes them worth knowing. ActID, in Shenzhen, launched in March: about 800 registered, roughly 300 agreed to license, and two productions have bought about 10 faces at $15 to $74 an episode. Rival New Claw set a price floor because users were undercutting each other. Its operations lead put it plainly: licensing a face is cheaper than hiring the person. Demand comes from microdramas — vertical, 90-second shows built for phones. China released roughly 128,000 in the first quarter of 2026; over 95% used AI in production. Huge appetite for faces, very little for paying performers. What built this market was THEFT. Two influencers accused ByteDance's drama platform of taking and altering their faces without consent; that drama had 40 million views before it was pulled. ByteDance says it removed 85,000+ videos with unauthorized AI faces and voices since January, and one Guangzhou court has heard roughly 700 face-theft cases in three years. The platforms pitch themselves as the fix: people take your face anyway, so you may as well be paid. The American version already exists. Twinnin launched in April, backed by Google and Nvidia. Actors pay $14.99 a year to list a digital twin; studios pay $499 to $1,200 a month to browse. Over 1,000 twins so far. On June 18, 2026, the Senate Judiciary Committee advanced the NO FAKES Act unanimously by voice vote — it would create a federal right to control how your likeness is used in a digital replica, with carve-outs for news, parody and criticism. Committee vote, not law. Tennessee's ELVIS Act has been in force since 2024, New York has regulated digital-replica contracts since 2025, and 45+ states have deepfake laws. The catch isn't the price. It's the paperwork. A Beijing lawyer who handles likeness cases told Rest of World that agencies routinely pay a few dozen dollars for facial data under terms so vague you can't tell who ends up using it, or for what. Once a face enters a marketplace, the person loses long-term control. No platform can stop face-swapping, scraping, or those images becoming future training data. A 21-year-old Shanghai model suing a brand over his face on products he never modeled says he'd never license his own — he'd have no way of knowing how AI might alter it. WHAT THIS MEANS FOR YOU: you are probably not licensing your face this week. But the machinery is being built now, the terms are being set now, and set cheap. If anyone asks to scan your face — an employer, a gig platform, a free headshot app, a loyalty program — ask three questions first. Who may use it. For how long. Can you revoke it. If the paperwork won't answer all three, that's your answer. Worth checking what your state gives you, too, since the federal bill isn't law yet. Somebody rented their face for $74. That's the going rate today, before any of this settles — and early rates stick. So: what's your number? What would someone have to pay you to license your face to an AI, permanently? Tell me in the comments. If this was useful, hit like, subscribe and tap the bell. Patreon link below. Keywords: rent your face to AI, AI likeness licensing, digital twin actors, NO FAKES Act, deepfake laws, biometric privacy, AI microdramas, face licensing, Twinnin, ELVIS Act, AI and your identity, AI news explained, artificial intelligence news 👇 Connect with The AI Guide: Website: www.theaiguide.ai YouTube: https://www.youtube.com/channel/UCamFJyTjb_kqUbNpVUVwQeg Patreon: www.patreon.com/theaiguide Facebook: www.facebook.com/davidtheaiguide Instagram: www.instagram.com/theaiguide LinkedIn: https://www.linkedin.com/company/the-ai-guide-on-youtube/ 🔗 Sources: [1] Rest of World — In China, people are renting out their faces to AI: https://restofworld.org/2026/china-ai-microdramas-face-licensing/ [2] Deadline — AI actors platform Twinnin divides opinion: https://deadline.com/2026/04/artificial-intelligence-twinnin-app-divides-film-tv-opinion-1236783395/ [3] Holland & Knight — Senate Judiciary and the NO FAKES Act: https://www.hklaw.com/en/insights/publications/2026/06/senate-judiciary-committee-advances-legislation-to-protect-name [4] Deadline — Twinnin funding round: https://deadline.com/2026/05/ai-plaform-twinnin-funding-round-3-million-signs-up-twins-1236882734/ #ArtificialIntelligence #AIDeepfakes #DigitalPrivacy #NOFAKESAct #TheAIGuide

The better AI gets, the MORE you'll pay for it. Dwarkesh Patel's math says one chip should rent for $250,000 a year. Here's a prediction almost nobody saw coming. While everyone argues about whether AI will get cheaper, analyst and podcaster Dwarkesh Patel laid out an argument in his essay "Why compute might get 10x more expensive in coming years" that flips the assumption on its head. Import AI, Jack Clark's newsletter, ran it as a lead item in issue #467 on August 3, 2026. The claim is simple and uncomfortable: the smarter these models get, the more expensive the hardware behind them becomes. Not less. More. The core of it is one thought experiment about pricing. Patel writes that if a true human-level software engineer could run on one H100-equivalent chip, then at current market rates for engineers, "that H100 should rent for over $250k a year. That's 15x today's spot prices." Read that again. Not fifteen percent higher. Fifteen TIMES higher. The logic is straightforward once you see it: hardware gets priced by what it can produce, not by what it costs to make. A chip that autocompletes code is worth a modest hourly rate. A chip that does a salaried professional's whole job is worth close to that salary. So what happens to your subscription when the model behind it crosses that line? The market is already drifting that way. Spot prices for high-end AI chips are reported up more than 40% from their February trough, and Google and Anthropic are reportedly paying SpaceX around $900 million a month for 110,000 GPUs, roughly twice the going spot rate. Supply cannot keep pace. Patel breaks the roughly 3x annual growth in lab compute into parts: about 1.4x from Moore's Law, 1.2x from new fabs, and 1.8x from AI grabbing a bigger share of leading-edge wafer allocation. That last lever is running out. By the end of 2027, AI is projected to move from around 60% of cutting-edge N3 capacity to 86%. Once you own nearly all of it, there's nothing left to take. Put demand next to that and the squeeze gets obvious. Anthropic's revenue has been roughly 10x-ing year over year and could reach $100 to $150 billion by the end of 2026. Continuing that curve means approaching a trillion dollars the year after, while lab compute grows only about 3x. Ten going into three doesn't fit. Something gives: margins rise, compute prices rise, or customers spend far more on inference. Patel admits margins in the mid-90s sound, in his words, quite crazy. That leaves price. He is careful about the counterarguments, which is what makes the piece worth taking seriously. Flood the world with AI engineers and the marginal value of an engineer drops, pulling that $250,000 figure down. And he concedes the long-run answer plainly: eventually robots will turn silica sand into computers and compute gets cheap again, but by then, in his phrasing, we'll be pretty deep into the singularity. Commenters raised one more: cheap Chinese open-weight models could undercut the whole premise. WHAT THIS MEANS FOR YOU: if you run a business on AI tools, or you've built a workflow around a twenty-dollar subscription, price stability is not something to plan around. Today's nearly free frontier intelligence is subsidized by the fact that these models still can't fully replace a skilled human. The moment that changes for your task, the pricing changes with it. Lock in annual plans where they make sense, watch the smaller open-weight models that run on hardware you control, and don't build a business whose margin depends on inference staying this cheap. So here's the question I'd love your take on: are AI prices heading up or down over the next two years, and would a 10x jump change how you use it? Drop a comment. Like, subscribe, and hit the bell to stay ahead on AI. Share this with someone who'd want to see it. Support the channel on Patreon (link below). Keywords: artificial intelligence, AI news, Dwarkesh Patel, compute prices, GPU prices, H100, Nvidia, AI costs, Anthropic, AI explained, AI for beginners, AI industry, data centers, AI economics, machine learning 👇 Connect with The AI Guide: Website: www.theaiguide.ai YouTube: https://www.youtube.com/channel/UCamFJyTjb_kqUbNpVUVwQeg Patreon: www.patreon.com/theaiguide Facebook: www.facebook.com/davidtheaiguide Instagram: www.instagram.com/theaiguide LinkedIn: https://www.linkedin.com/company/the-ai-guide-on-youtube/ 🔗 Sources: [1] Import AI 467 — Jack Clark, Aug 3, 2026: https://importai.substack.com/p/import-ai-467-self-sustaining-ai [2] Why compute might get 10x more expensive — Dwarkesh Patel: https://www.dwarkesh.com/p/why-compute-might-get-10x-more-expensive [3] AI Compute Could Get 10-15x Pricier — AI Weekly: https://aiweekly.co/alerts/dwarkesh-patel-argues-ai-compute-could-get-10-15x-pricier [4] Could AI Compute Get 10x More Expensive? — metir: https://www.metirai.com/blog/why-ai-compute-more-expensive-scaling-economics-2026 #ArtificialIntelligence #AINews #DwarkeshPatel #GPUPrices #TheAIGuide