
The voice-to-text app that built its own speech model
Most dictation apps are a wrapper around someone else's speech model. Wispr Flow trained its own.
It's called Canto, and it was built for the way people actually talk to AI: thinking out loud, changing direction mid-sentence, half-whispering in an open office.
Wispr tested it on 10 hours of real dictations from more than 2,300 speakers. Canto had the lowest word error rate of every model tested, including models from Google, OpenAI, AssemblyAI and Deepgram. That means your prompts to ChatGPT, Claude and Cursor come through the way you said them.
Talk naturally in any app and Wispr Flow types clean, formatted text, 4x faster than typing.
⚖️ POLICY & REGULATION
GAO gets four AI chatbots to draft U.S. attack plans: In a report released Wednesday, the Government Accountability Office said ChatGPT, Claude, Gemini and Grok each produced plans for a terrorist attack in the United States in under 12 hours of testing, CBS News and The Associated Press reported. Sen. Maggie Hassan, D-N.H., requested the 61-page report. Two platforms produced plans for improvised explosive devices meant to cause mass casualties. Two others planned drone or firearm attacks on the electrical grid. GAO said the information was largely not novel, but the tools compressed weeks of planning into hours. The stakes: Hassan becomes chair of the Senate Homeland Security Committee in January and told the AP one of its first hearings will cover AI risks. She also plans legislation to tighten safety requirements. Anthropic, Google, OpenAI and SpaceXAI answered in the report; Google said the testing used earlier models. Link
Democrats press DOJ on its OpenAI copyright brief: Congressional Democrats are circulating a letter to Attorney General Todd Blanche asking whether the Justice Department was biased when it backed OpenAI and Microsoft in The New York Times' copyright suit, Axios reported Tuesday. Rep. Chellie Pingree, D-Maine, is circulating it. The letter asks for the drafts behind the department's Sept. 1 statement of interest and the names of any outside groups that supplied legal analysis. The lawmakers say they are "alarmed" that the Copyright Office and the Patent and Trademark Office were not consulted. The stakes: The statement argued that training AI on news articles is fair use. It does not bind the court, but it is the administration's stated position on the training question every AI developer faces. No Republican has signed the letter. Next: Oversight letter. Answers requested by Nov. 30. Link
Democrats rally for a national AI data center moratorium: Reps. Jim McGovern, D-Mass., and Chellie Pingree, D-Maine, joined Reps.-elect Adam Hamawy of New Jersey and Will Lawrence of Michigan outside the Capitol on Tuesday to call for a nationwide moratorium on new AI data centers, the Missouri Independent reported. Food & Water Watch hosted the event. The pause would last until Congress enacts AI safeguards, as in S. 4214 from Sen. Bernie Sanders, I-Vt., and H.R. 9442 from Rep. Alexandria Ocasio-Cortez, D-N.Y. McGovern, the top Democrat on the House Rules Committee, is an original cosponsor of the House bill. The stakes: Neither bill has moved since its referral. House Minority Leader Hakeem Jeffries suggested in July that he does not back a national moratorium, NOTUS reported, so the thing to watch is whether leadership lets it reach a vote next year. Next: In committee. Track S. 4214 on Docket Room. Link
This week on Docket Room
STATES
Oregon sues OpenAI for $2 billion over ChatGPT and kids: Oregon Attorney General Dan Rayfield sued OpenAI on Tuesday in Multnomah County Circuit Court, seeking at least $2.035 billion under the state's Unlawful Trade Practices Act, Quartz reported. The complaint says OpenAI misled parents about ChatGPT's safety for children. The state wants an injunction barring OpenAI from misrepresenting the safety and age-appropriateness of ChatGPT and ChatGPT for Teens, Engadget reported. It cites OpenAI's imperfect age prediction. OpenAI has added parental controls and did not answer Engadget before publication. The stakes: Oregon joins Florida, which sued in June, and the eight states that filed coordinated suits in October. Unlike California's Nov. 16 suit over AI agent failures, this is a consumer protection case about minors. In our assessment, a $2 billion demand raises the price of settling the state cases. Link
Texas resumes data center permits before ERCOT reports: The Texas Commission on Environmental Quality has resumed reviewing data center permits after Gov. Greg Abbott's office told it Friday that the grid operator had finished gathering information, The Texas Tribune reported Tuesday. Abbott had told the agency on Sept. 21 to wait until ERCOT completed its review. ERCOT said Nov. 5 that 93% of 465 projects had answered its audit. It plans to report findings to the Public Utility Commission on Dec. 10. Public Citizen said the agency is issuing permits before those results are public. The stakes: Abbott lost re-election Nov. 3 to Democrat Gina Hinojosa, who campaigned on keeping approvals paused until the Legislature acts. In our assessment, permits issued before she takes office would be hard to unwind. Next: Regulatory stage. ERCOT reports to the PUC Dec. 10; the PUC discusses the findings Dec. 17. Link
San Jose and Salt Lake City adopt data center pauses: San Jose's City Council unanimously adopted a moratorium Tuesday on land use approvals for data centers and other large energy users, pausing 12 projects under review, San José Spotlight reported. It lasts until the city's new standards take effect. Salt Lake City's council unanimously approved a 180-day pause the same night, the maximum state law allows, KUTV reported. Council staff told Fox 13 that a vested Northpoint application could still proceed. San Francisco supervisors unanimously extended their pause to next September, the San Francisco Standard reported. Pasquotank County, North Carolina, adopted a 12-month moratorium Monday, the Albemarle Observer reported. County staff said state law likely bars banning a whole category of development. The stakes: Each of the four is a pause tied to permanent rules, not a ban. San Jose's holds 12 live projects, so the Dec. 8 standards decide what those developers must meet on energy, water, air and noise. Next: Council stage. San Jose takes up its standards Dec. 8 as an urgency ordinance. Link
🛡️ SAFETY & SECURITY
Hilton and House sponsors press for a DEFIANCE Act vote: Paris Hilton joined Reps. Alexandria Ocasio-Cortez, D-N.Y., and Laurel Lee, R-Fla., at the Capitol on Tuesday to urge a House floor vote on the DEFIANCE Act, CBS News reported. The bill would let victims of sexually explicit deepfakes sue the people who create or share them. The Senate passed S. 1837 by unanimous consent in January. Ocasio-Cortez said it has sat untouched in the House for 10 months. Hilton appealed directly to Speaker Mike Johnson, Majority Leader Steve Scalise and Judiciary Chairman Jim Jordan. The stakes: The Take It Down Act already requires platforms to remove these images. DEFIANCE adds a civil damages claim against the creators and sharers. A Senate-passed bill the House does not take up before January has to start over in the new Congress. Next: Passed the Senate; House companion H.R. 3562 is in Judiciary. Track S. 1837 on Docket Room. Link
New York AG opens a unit to enforce the RAISE Act: New York Attorney General Letitia James created an AI Safety, Standards, Enforcement and Transparency Unit on Tuesday to enforce the RAISE Act, her office announced. The ASSET Unit sits in the Economic Justice Division, works with the Bureau of Internet and Technology and is led by Cristy Phillips. The RAISE Act, S 8828 (Chapter 96 of 2026), takes effect Jan. 1, 2027. It requires large frontier AI developers to publish safety frameworks and report critical incidents to the state. The unit will also cover privacy, cybersecurity, scams and online safety. The stakes: Registration for frontier developers opened Nov. 10. A named enforcement team means the first incident reports under the law arrive at an office that can bring penalty actions. The AG's office says it is still building the team. Next: In force Jan. 1, 2027. Link
📜 DISPATCH PRO
The working layer under this briefing is The Docket, in Dispatch Pro
What moved in all 50 states since the last issue, which AI bills actually matter this week, and exactly what they would make you do, by when: obligations, deadlines and penalties pulled from the bill text and verified, similar bills in other states, every hearing and effective date we have reported, plus a Docket Room Pro seat ($49/mo on its own).
Today for Pro readers: 5 AI bills moved since the last issue, 0 rated Act; this week's three; The Bill on US S 3741 with 4 obligations, deadlines and penalties pulled from the text; 1 more AI hearing this week and 50 dated actions in the next 30 days.
$29/mo or $290/yr after the trial, less than the Docket Room seat alone. Cancel anytime. Team licenses on request.
Know someone who should be reading this?
Digitize Dispatch is read by lobbyists, government affairs, policy, and regulatory professionals tracking AI legislation, regulation, and policy developments. One sponsor per issue, at the top. See rates and sponsor the Dispatch.
Have feedback? Send us an email.



