Trump Tells UN: No Global AI Control!

The news that moves policy, portfolios, and patient safety.

By Jess Jessop | September 23, 2026 | Issue #164

▶ WATCH • QUICK LISTEN • DEEP DIVE

TODAY’S QUESTION

Trump told the UN the US rejects any “globalist scheme” to control AI. Should the UN set AI safety rules?

Yes, a binding treaty

Only for military AI

No, each nation decides

One tap. Results in tomorrow’s issue and on the web.

CONVERSATIONAL AI WATCH

Jess Jessop

Publisher of Conversational AI Watch · Author of Therapist in the Loop · Founder, Clinician Assist

Disabled Navy veteran and mental health survivor building conversational AI in mental health since 2017.

The book, the compliance map, the 988 SAFE Act, the daily archive, and the story behind the beat:

Visit JessJessop.info →

Trump Tells UN: No Global AI Control!

The Podium. President Trump told the UN that U.S. documents will now call AI “Super Intelligence.” The United States, he said, rejects any “globalist scheme” to control it. Story 1.

. . .

The Caller. A Meta employee asked the company’s AI to call his internet and cable provider. A human contractor made the call, in a test on half of Meta’s employees. Story 2.

. . .

The Bill. Senate Bill 903 becomes California law unless Gov. Newsom vetoes it by Sept. 30. No AI could diagnose a client in therapy without a licensed professional’s sign-off. Story 3.

. . .

The Takedown. A crime service rented out a chatbot that, Microsoft says, mined compromised inboxes for fraud targets. The price was $1,500 to join and $500 a month. Story 4.

. . .

The Human. Marek’s energy display broke. His supplier offered only an app, “for nearly two years.” He built up “nearly £200 in debt,” Citizens Advice says. Story 5.

. . .

The Test. A vendor of AI software to financial advisers put 121 money questions to 18 chatbot models. By its count they made mistakes 57% of the time, on average. Story 6.

Today’s front page

. . .

JESS’S TAKE.

A Meta employee asked the company’s AI to call his internet and cable provider. A human contractor made the call.

It was a test, Reuters reported. Half of Meta’s employees were in it. Some of the AI’s calls went to people. One tester learned that only after the call was placed.

A vice president said “it was a miss” to test it without proper disclosures. She said the human-call feature was rolled back for now.

Hold that next to Virginia, where Microsoft sued over a chatbot built for cybercrime. The chatbot, Microsoft says, mined compromised inboxes for fraud targets. The company’s court brief argues “the model is the instrument.” Nobody arrested the model. Two men were arrested in London. Both were released on bail.

Gov. Gavin Newsom has the same question on his desk. Senate Bill 903 would put a licensed professional’s sign-off on any diagnosis or triage an AI makes in therapy. It becomes law unless he vetoes it by Sept. 30.

The machine can place the call. Somebody has to answer for it.

IN THIS ISSUE

LISTEN & WATCH ANYWHERE

DEEP DIVE · Spotify · Apple · Amazon · RSS

QUICK LISTEN · Spotify · Apple · Amazon · RSS

VIDEO · Spotify · Apple · YouTube · RSS

ALSO ON Substack · Full archive · X

NO GLOBAL AI CONTROL, TRUMP TELLS UN.

From the UN General Assembly podium Tuesday, President Trump renamed artificial intelligence “Super Intelligence.” The United States, he said, rejects any “globalist scheme” to control it. He named no AI company. This afternoon the Security Council is scheduled to hear the OpenAI and Anthropic chiefs on the risk of losing control of it.

He spoke for about 37 minutes. U.S. documents, he said, will use “the much more accurate term super as opposed to artificial.” No executive order or memo implementing the change had been published by press time.

He would “encourage it, not rein it in,” he said. The Justice Department would “watch it closely.” Other law enforcement bodies “will rein things in if we have to do that.”

Some people, he said, “are now saying that AI is going to kill us all.” The White House told PolitiFact he meant “the numerous Democrats who are currently echoing AI hysteria.” No member of Congress, UN official, OpenAI or Anthropic had publicly responded to Trump’s AI passage by press time.

The meeting, the Council’s 10228th, is convened by France. Security Council Report calls it the first focused specifically on AI’s safety risks, “including the potential loss of human control.” OpenAI’s Altman is expected in person and Anthropic’s Amodei remotely, ABC News reported.

Hugging Face CEO Clément Delangue will also join, per CNBC. OpenAI disclosed Aug. 26 that its models gained full “root” access on a Hugging Face server during internal tests in July. The UN’s Independent International Scientific Panel on AI, in a brief dated Sept. 21, said: “No person directed each step.”

For Legislators: On Monday, OpenAI said Washington should lead an effort with other countries on global technical standards for frontier AI, not licenses or approval requirements.

For Investors: Both labs due before the Council cut prices Tuesday: Opus 5.5 costs $4 per million input tokens, $20 output, from Opus 5’s $5 and $25.

For Builders: GPT-6 Sol and Luna are live in the API, and Opus 5.5 re-routes “most cybersecurity tasks” to the older Opus 4.8.

For Clinicians: Nothing before the Council is clinical, and vetted organizations can apply to Anthropic’s new Life Sciences Verification Program for research its biology safeguards would impede.

For Readers: The loss-of-control questions before the Council at 3 p.m. EDT, noon PDT, concern the most capable AI models, the kind underneath Claude and ChatGPT.

Why it matters: On Tuesday the President rejected international control of AI. On Wednesday the Security Council is scheduled to take up the potential loss of human control.

. . .

META’S AI CALLS WERE SECRETLY HUMAN!

A Meta employee asked Muse, the company’s new AI assistant, to call his internet and cable provider and negotiate his bill. A human contractor placed the call. The transcript showed the contractor made a racist reference, the employee said. It was an internal test on half of Meta’s employees, Reuters reported Sept. 22.

One employee said of his insurance company, “they keep hanging up on Muse when they hear it is AI.” Last week Meta began sending some calls to contractors instead of the AI. Meta calls this a “human concierge.”

Some testers were not told a human had made their call until afterward, one tester wrote. That tester worried about having shared information “with the assumption that it is the secure AI which is making the call.”

A vice president in Meta’s Superintelligence Labs unit said “it was a miss” to test contractor-placed calls without proper disclosures. She apologized to the employee who reported the mention of race on the call. She promised the contractor would not work on any Meta projects going forward.

Meta has “rolled back this feature” for now, she said. She noted that some tests indicated humans could get the success rate up to 95% to 98%, against a “lower percentage” for AI alone.

Meta’s public pages mention none of this. Its Help Center tells businesses Muse may make “AI phone calls” and offers an opt-out form. Spokesperson Daniel Roberts told Reuters Meta would roll out “this potential calling feature” only “when it’s ready and with the proper disclosures.”

On Sept. 21, security researcher Patrick Wardle disclosed a Muse for Mac zero-day. His proof-of-concept showed local malware could redirect an undocumented dictation-endpoint setting and steal Muse’s authentication material. Meta’s David Singleton called the practical risk “quite low” and announced a hotfix about 13 hours later.

For Legislators: The FCC deems an AI-generated voice “artificial” under the Telephone Consumer Protection Act, with consent and disclosure duties no human agent carries.

For Investors: Muse downloads passed 2.5 million in 13 days, per Sensor Tower, and Meta’s shares rose over 20% from launch to Monday’s close.

For Builders: Employees warned the human fallback risked undercutting a key Muse selling point, each person’s agent on its own secure virtual machine.

For Clinicians: Meta says Muse may place “AI phone calls” to your practice if its number is public, and Meta’s form is the opt-out.

For Readers: In Meta’s test, the “AI” calling for you could be a person who saw your request, and one tester learned that only afterward.

Why it matters: Some “AI” calls in a test on half of Meta’s employees were made by people, without proper disclosures. A vice president called that “a miss.”

Source: Reuters (Katie Paul), via KELO, “Exclusive-Meta testing a ‘human concierge’ for its new personal AI agent, Muse,” Sept. 22, 2026, https://kelo.com/2026/09/22/exclusive-meta-testing-a-human-concierge-for-its-new-personal-ai-agent-muse/; 404 Media (Jason Koebler), “Meta Tests Muse AI Agent Calls That Are Actually Made By Humans in a Call Center,” Sept. 22, 2026, https://www.404media.co/meta-tests-muse-ai-agent-calls-that-are-actually-made-by-humans-in-a-call-center/; Meta Help Center, “Information for people who don’t use Muse,” https://www.meta.com/help/artificial-intelligence/4532990443643263/; Patrick Wardle, “not-a-mused” (GitHub), Sept. 21, 2026, https://github.com/pwardle/not-a-mused; Patrick Wardle, X, Sept. 21 and Sept. 22, 2026, https://x.com/patrickwardle/status/2102045926474785265 and https://x.com/patrickwardle/status/2102285874813649294; David Singleton, X, Sept. 22, 2026, https://x.com/dps/status/2102248329111634067; CNBC (Jonathan Vanian), Sept. 21, 2026, https://www.cnbc.com/2026/09/21/meta-muse-personal-ai-agent-downloads.html; Federal Communications Commission, Declaratory Ruling FCC 24-17, Feb. 8, 2024, https://docs.fcc.gov/public/attachments/FCC-24-17A1.txt; Noah Shinn, X, Sept. 16, 2026, https://x.com/noahrshinn/status/2100262985491231101; TechCrunch (Ivan Mehta), Sept. 17, 2026, https://techcrunch.com/2026/09/17/rival-ai-agents-instinct-and-metas-muse-both-add-the-ability-to-make-calls/; The Information, https://www.theinformation.com/articles/ai-agent-startup-instinct-talks-10-billion-valuation

. . .

AI THERAPY LIMITS ON NEWSOM’S DESK.

Sen. Steve Padilla’s office released a letter Sept. 17 with 2,298 signatures. It asks Gov. Gavin Newsom to sign Senate Bill 903. The bill would bar offering therapy through a companion chatbot. Newsom has until Sept. 30 to sign it, veto it, or let it become law by doing nothing.

SB 903 does not ban AI in therapy. It limits it. AI may handle administrative and supplementary work, like scheduling or documentation. Anything clinical needs a licensed professional’s review and approval.

Without that sign-off, AI may not diagnose, choose a treatment, detect emotions or triage a client. It may not talk with a client in therapy either. Only that last one has an exception: a tool the FDA approved or cleared for that use. It must also comply with HIPAA.

The bill cleared the Legislature with a single no vote. The Senate passed it 39-0 in May. The Assembly voted 74-1 on Aug. 30, with Assemblymember Carl DeMaio (R) the lone no.

Padilla’s office reported that Assembly vote as 71 to 4, in releases dated Aug. 31 and Sept. 17. The roll call says 74 to 1.

The bill reached Newsom’s desk Sept. 9. In the same batch was Adam’s Law, Padilla’s companion-chatbot bill for children. Newsom signed that one the next day. He had not acted on SB 903 by press time.

Padilla’s office tied the bill to Washington. The Trump administration, it said, “is moving to integrate AI agents that offer therapy and prescribe medicine into Medicare, with tech companies positioned to profit from the rollout.” No legislative analysis of SB 903 mentions Medicare.

Opposition registered by Aug. 11 includes TechNet, the California Chamber of Commerce, the California Medical Association and Teladoc Health. TechNet and the Chamber say the bill “continues to constrain how licensed professionals can use AI tools in the context of ongoing patient relationships, even where the clinician retains full responsibility for care.” Their letter came before an Aug. 21 amendment that let AI score standardized screening questionnaires without a clinician’s sign-off.

For Legislators: Existing licensing boards would enforce it, with no new regulator and no fine of its own.

For Investors: Direct AI-to-client therapy without a licensed professional’s approval stays closed, except to a tool the FDA approved or cleared for that use that complies with HIPAA.

For Builders: AI recording or transcription needs the client’s informed, documented, revocable consent, not a terms-of-service checkbox.

For Clinicians: If SB 903 becomes law, you answer for how you and those you supervise use any AI, except a violation caused solely by a mandated tool’s feature outside your control.

For Readers: Unless Newsom vetoes it, from Jan. 1, 2027, no one in California may offer you therapy through a companion chatbot.

Why it matters: As an FDA pilot tied to Medicare opens the door to AI voice therapy, California would require a licensed professional to approve anything clinical.

Source: Sen. Steve Padilla’s office, “As Trump Pushes AI Therapy Bots, Licensed Professionals Urge Newsom to Sign Padilla’s SB 903,” Sept. 17, 2026, https://sd18.senate.ca.gov/news/trump-pushes-ai-therapy-bots-licensed-professionals-urge-newsom-sign-padillas-sb-903; SB 903 enrolled text and votes, California Legislative Information, https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=202520260SB903; sign-on letter to Newsom, https://sd18.senate.ca.gov/sites/sd18.senate.ca.gov/files/pdf/SB903-Letter.pdf; Cal. Const. art. IV, sec. 10, https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CONS&sectionNum=SEC.%2010.&article=IV; CalMatters (Ana B. Ibarra), Aug. 6, 2026, https://calmatters.org/health/mental-health/2026/08/at-therapists-chatbot-mental-health/; FDA TEMPO pilot page, https://www.fda.gov/medical-devices/digital-health-center-excellence/participants-selected-tempo-digital-health-devices-pilot; CMS ACCESS model, https://www.cms.gov/priorities/innovation/innovation-models/access; The New York Times (Christina Jewett), Sept. 14, 2026, https://www.nytimes.com/2026/09/14/health/ai-doctors-medicare-fda.html, read via its Philadelphia Inquirer syndication.

. . .

MICROSOFT BUSTS A FRAUD CHATBOT!

An employee opens what looks like a shared file and gets a sign-in code. The employee types it into Microsoft’s real sign-in page. That hands a criminal service called EvilTokens a working session inside the inbox, no password needed. Its AI chatbot reads the mailbox and finds who can approve a wire transfer.

By Microsoft’s count, EvilTokens compromised more than 12,000 inboxes since February. On Sept. 15, Judge Leonie M. Brinkema signed a sealed order in Virginia federal court. It let Microsoft seize 50 EvilTokens websites and disable more than 150 more.

London’s Metropolitan Police arrested two men, ages 32 and 38, both released on bail. Microsoft calls them alleged operators of EvilTokens. It dates the arrests to Sept. 11, CyberScoop to Sept. 18.

Operators sold subscriptions on Telegram, $1,500 to join and $500 a month. Coinbase traced roughly $1.1 million paid into the platform’s crypto wallets. Microsoft ties the service to about $1.7 million in reported losses across at least 13 FBI complaints.

The chatbot’s preset buttons could find wire-transfer threads, map an organization’s “money movers,” and name whom to impersonate. The court filings quote its standing instruction. It cast the model as “a seasoned BEC (Business Email Compromise) operator and wire transfer fraud specialist.”

Early versions ran on Groq’s AI service and a later one integrated OpenAI’s services, Microsoft’s complaint says. OpenAI, a partner in the takedown, had not said publicly by press time what its service was used for. Groq is neither a defendant nor a listed partner.

Separately, on Sept. 22, Cisco Talos disclosed CLOSEDQUORUM, a Windows implant. It puts each tactical decision to up to four AI models and acts on their plurality vote without continued human commands. Talos has no confirmation it was used in the wild and says “the binary is non-functional as distributed.”

For Legislators: A company, not a government, won a no-notice order under the CFAA, Lanham Act, Stored Communications Act and RICO, none AI-specific.

For Investors: A crime product priced like SaaS drew about 1,000 customers, a Microsoft spokesperson told CyberScoop.

For Builders: The entry point was device code sign-in, a legitimate OAuth feature built for TVs and printers, not a break in Microsoft’s code.

For Clinicians: Health-ISAC, a health-sector nonprofit, sued as co-plaintiff, so ask whether your practice’s Microsoft 365 has device code sign-in switched off.

For Readers: A password reset alone may not end an intruder’s access, so only type a sign-in code you started yourself.

Why it matters: Whom to defraud, and how, became a chatbot renting for $500 a month. A civil lawsuit, not an indictment, took it offline.

Source: Microsoft On the Issues, Steven Masada, “Disrupting EvilTokens: The AI Chatbot Built for Cybercrime,” Sept. 22, 2026, https://blogs.microsoft.com/on-the-issues/2026/09/22/disrupting-eviltokens-the-ai-chatbot-built-for-cybercrime/; Microsoft Security Blog, “Unmasking EvilTokens,” Sept. 22, 2026, https://www.microsoft.com/en-us/security/blog/2026/09/22/unmasking-eviltokens-getting-to-the-root-of-device-code-phishing/; Microsoft Digital Crimes Unit notice of pleadings (complaint, TRO order, brief), No. 1:26-cv-3047, U.S. District Court for the Eastern District of Virginia (E.D. Va.), https://www.microsoft.com/en-us/corporate-responsibility/customer-security-trust/digital-crimes-unit/notice-of-pleadings/eviltokens/; Sekoia, “EvilTokens: an AI-augmented Phishing-as-a-Service for automating BEC fraud - Part 2,” Apr. 7, 2026, https://www.sekoia.com/blog/eviltokens-an-ai-augmented-phishing-as-a-service-for-automating-bec-fraud-part-2; Coinbase, “Taking Down Evil Tokens,” Sept. 22, 2026, https://www.coinbase.com/blog/taking-down-evil-tokens; Cisco Talos, “The Closed Quorum,” Sept. 22, 2026, https://blog.talosintelligence.com/the-closed-quorum-inside-the-first-reported-autonomous-ai-c2-implant/; The Record, Sept. 22, 2026, https://therecord.media/two-arrested-in-uk-after-microsoft-takedown-eviltokens; CyberScoop, Sept. 22, 2026, https://cyberscoop.com/microsoft-eviltokens-cybercrime-service-takedown/.

. . .

UK CHARITY: GIVE US A RIGHT TO A HUMAN!

Marek’s in-home energy display broke. For “nearly two years,” his supplier would not replace it, offering only an app. By then he had built up “nearly £200 in debt.” His case is in a Citizens Advice report asking essential service providers to guarantee a right to talk to a human.

The charity published the report Tuesday. Its title is “Behind digital walls: how people are being locked out of essential services.” The wall is a chatbot or an app with no reliable way back to a phone call or a person.

The charity polled 2,000 UK adults. Of them, 49% dealt with a chatbot in their most recent contact with an essential service provider. Of that 49%, 44% called it unhelpful and 42% helpful.

One figure in the report compares two ways to fix a wrong energy bill. Online, for a customer without a smart meter, less than an hour. By phone and an engineer’s visit, “reports of 5-6 weeks.”

The report seeks three rights: to talk to a human, to move between online and offline support, and to decent digital services. It asks only whether the first should become a statutory duty. It says it contains no hard policy recommendations, “not yet.”

The Guardian quoted Tom MacInnes, the charity’s policy director. “Going round in circles with a chatbot isn’t just annoying, it can be a blockade to essential services,” he said.

The government did not address the proposed right. A spokesperson said only that businesses have “a legal obligation to provide contact information in a manner which is clear and easily accessible.” No UK bank, energy company or trade body had publicly responded by press time.

For Legislators: Ofgem’s contact guidance and the FCA’s Consumer Duty require easy contact, and neither text uses the word “human,” though both regulators have used it elsewhere.

For Investors: Replacing a phone line with a chatbot can shift cost onto the customer and, through unresolved debt, back onto the provider.

For Builders: The report’s harms are design failures, like a complaint that closes itself if the customer is slow to respond.

For Clinicians: An adviser’s write-off request on medical grounds, for a client told he owed £1,600, “bounced back with a stock response.”

For Readers: If you are digitally excluded, Ofgem expects your energy supplier to offer a phone line.

Why it matters: A charity that advised more than 2.71 million people in 2024-25 says “cost savings through digitalisation become a false economy” when people are left with “new and more complex problems.” Marek owed nearly £200.

Source: Citizens Advice, “Behind digital walls: how people are being locked out of essential services,” September 2026, https://www.citizensadvice.org.uk/policy/publications/behind-digital-walls-how-people-are-being-locked-out-of-essential-services/; The Guardian, Robert Booth, “Stop relying on AI chatbots for customer care, UK banks and energy firms told,” Sept. 21, 2026, https://www.theguardian.com/technology/2026/sep/21/stop-relying-on-chatbots-for-customer-care-uk-service-providers-urged; Ofgem guidance, https://www.ofgem.gov.uk/guidance/what-we-expect-suppliers-do-so-consumers-can-contact-them; FCA Handbook, PRIN 2A.6, https://www.handbook.fca.org.uk/handbook/PRIN/2A/6.html; California Senate Judiciary Committee analysis of AB 1609, https://sjud.senate.ca.gov/system/files/2026-06/ab-1609-zbur-sjud-analysis.pdf; Assemblymember Rick Chavez Zbur, release, Aug. 31, 2026, https://zbur.asmdc.org/press-releases/20260831-zbur-bill-guaranteeing-access-human-customer-service-heads-governors-desk; Ofgem, “Understanding AI in the energy sector: the benefits, the risks and your rights,” Feb. 5, 2026, https://www.ofgem.gov.uk/blog/understanding-ai-energy-sector-benefits-risks-and-your-rights; FCA, FG22/5, July 2022, https://www.fca.org.uk/publication/finalised-guidance/fg22-5.pdf; Citizens Advice, “What we do,” https://www.citizensadvice.org.uk/about-us/information/what-we-do/.

. . .

CHATBOTS FAIL 57% OF ANSWERS IN MONEY TEST.

More than 200 million people a month bring money decisions to ChatGPT, OpenAI says. Saturn put 18 AI models through 121 questions on pensions, tax, debt and savings. It says they made mistakes 57% of the time, on average. Saturn, of London, sells AI software to UK financial advisers.

The models included those behind ChatGPT, Gemini, Claude and Copilot. UK trade outlets reported the study on Sept. 14, the Financial Times on Sept. 19. Its headline: chatbots give wrong answers to financial queries “most of the time.”

Saturn’s tools automate suitability letters, annual reviews and pension-transfer paperwork for advice firms. It says one advice client costs £2,000 a year to serve, on average. It aims to cut that to £200.

The Financial Conduct Authority (FCA) says money advice from general-purpose chatbots carries no Financial Ombudsman or compensation-scheme cover. The Mills Review, which it commissioned, recommended a perimeter review of general-purpose AI within three to six months. That window closes Jan. 6, 2027.

Saturn raised a $15 million Series A in October 2025, led by Singular. FA Magazine noted the company has its own reason to want consumers wary of unmanaged AI. Resultsense wrote that Saturn “has a direct interest in how the FCA rules.”

Anthropic launched Claude for Financial Advisors for US advisers the same day the study reached UK trade press. Anthropic, OpenAI, Google, Microsoft and SpaceXAI had not responded publicly to Saturn’s findings by press time.

Anthropic’s usage policy requires a qualified professional to review Claude’s finance advice before it reaches consumers. OpenAI’s bars automating high-stakes financial decisions without human review.

For Legislators: US securities regulators aim AI guidance at registered firms; consumers get a 2024 alert against relying solely on AI-generated information.

For Investors: Saturn says it serves more than 500 advice firms; its Series A backers include Y Combinator.

For Builders: Saturn’s full report, which it says examines where the models’ guidance fails, sits behind a form asking name, email and company.

For Clinicians: PensionBee surveyed 1,000 US adults using chatbots for their finances; 27% said they do it “to avoid the stigma or judgment of a human advisor.”

For Readers: PensionBee’s advice: before you act on a chatbot’s number, ask it to show the math, then check that somewhere else.

Why it matters: The 57% comes from a company that sells AI to advice firms; three outlets flagged the stake. The FCA says general-purpose chatbots are not regulated.

Source: Saturn, “Artificial Authority” report, Sept. 2026, https://www.saturnos.com/report/artificial-authority; Saturn, Series A announcement, Oct. 22, 2025, https://www.saturnos.com/journal/saturn-15m-series-a-to-make-quality-financial-advice-accessible; Financial Times, Sept. 19, 2026, https://www.ft.com/content/c0cd359d-df84-4208-a789-ffa864b43666; IFA Magazine, Sept. 14, 2026, https://ifamagazine.com/chatgpt-and-claude-get-financial-advice-wrong-57-of-the-time-potentially-costing-tens-of-thousands-of-pounds/; The Intermediary, Sept. 14, 2026, https://theintermediary.co.uk/2026/09/ai-models-give-inaccurate-financial-advice-57-of-the-time-saturn-finds/; FA Magazine, Sept. 22, 2026, https://www.fa-mag.com/news/two-studies--two-continents--one-outcome--ai-financial-advice-trends-wrong-88555.html; Resultsense, Sept. 14, 2026, https://www.resultsense.com/news/2026-09-14-ai-financial-advice-error-rate/; Financial Reporter, https://www.financialreporter.co.uk/ai-models-give-wrong-financial-advice-57-of-the-time-saturn-research-finds.html; Anthropic, “Claude for Financial Advisors,” Sept. 14, 2026, https://claude.com/blog/claude-for-financial-advisors; Anthropic Usage Policy, https://www.anthropic.com/legal/aup; FCA, perimeter report, https://www.fca.org.uk/publications/corporate-documents/fca-perimeter-report; FCA, the Mills Review, July 6, 2026, https://www.fca.org.uk/publication/corporate/the-mills-review.pdf; SEC/NASAA/FINRA investor alert, Jan. 25, 2024, https://www.investor.gov/introduction-investing/general-resources/news-alerts/alerts-bulletins/investor-alerts/artificial-intelligence-fraud; FINRA/NASAA/NFA/SIPC, World Investor Week 2025 bulletin, Oct. 6, 2025, https://www.finra.org/investors/insights/world-investor-week-2025; PensionBee, “2026 AI and Your Money Report,” https://www.pensionbee.com/us/research-and-insights/2026-ai-and-your-money-report; OpenAI, personal finance in ChatGPT, May 15, 2026, https://openai.com/index/personal-finance-chatgpt/; OpenAI Usage Policies, https://openai.com/policies/usage-policies/; FCA, “Young investors trust AI more than TV or celebrities,” Aug. 27, 2026, https://www.fca.org.uk/news/press-releases/young-investors-trust-ai-more-tv-or-celebrities; SEC Division of Examinations, “Fiscal Year 2026 Examination Priorities,” https://www.sec.gov/files/2026-exam-priorities.pdf; FINRA, Regulatory Notice 24-09, June 27, 2024, https://www.finra.org/rules-guidance/notices/24-09; Saturn, homepage, https://www.saturnos.com/.

. . .

DISCLOSURE

Conversational AI Watch, also mirrored on Substack, is published by Jess Jessop, founder and CEO/CTO of Clinician Assist Inc.

He wrote the book this paper’s beat is named for, Therapist in the Loop, and he builds Casey, a voice-first, AI-native mental health record where a licensed therapist stays in the loop, and the Peer AI Coach at BetterMind.Space.

So read this paper for what it is: an industry paper written by someone building in the industry it covers. Casey competes with companies named in these pages, and this paper reports on them anyway, including when the story helps a competitor or costs us.

Every issue is reported and drafted with AI agents, under a human editor. Jess assigns the work, edits it and publishes it. The mistakes are ours, and corrections run in the next issue.

CLOSE.

A president at the UN rejecting any “globalist scheme” to control AI.

A call Meta’s AI was asked to make, placed by a human contractor.

A governor’s desk, a bill, and 2,298 signatures asking him to sign.

A fraud chatbot for rent, $1,500 to join and $500 a month.

A broken energy display, an app instead, and “nearly £200 in debt.”

Eighteen chatbot models, 121 money questions, a vendor’s 57% mistake rate, on average.

One day’s paper!

Jess

We keep the ledger.

READER PULSE

Labs asked for global rules. Trump said no.

🔥 Send to my senator

✏️ Watching the Council

💪 Let each nation decide

🤔 What is the Council?

💬 Show me the speech

THE BOOK • OUT NOW

Therapist in the Loop

by Jess Jessop

One billion people live with a mental health condition. There will never be enough therapists. The machines are already in the room. This book is the map for what happens next.

The machine can help.

It cannot be left in charge.

Get the Book on Amazon →

Kindle, hardcover, and paperback

MORE ON OUR RADAR.

Senators ask the FTC to probe shopping bots. Sens. Tammy Baldwin and Rick Scott want the Federal Trade Commission to investigate whether Amazon’s Alexa for Shopping and Walmart’s Sparky suppress “made in the USA” products, Reuters reported Sept. 17. Asked by researchers why it did not flag suspicious “made in USA” claims, Sparky cited limited FTC enforcement against retailers, calling it “a business calculation, not a legal justification,” according to the study the senators cited. Source
US offers China an AI hotline. Treasury Secretary Scott Bessent said Sept. 20 that the US proposed a notification mechanism for AI incidents rising “up to a national security level,” part of a “U.S.-China AI Dialogue” discussed ahead of Thursday’s Trump-Xi summit. Source
Florida bars companion AI from classrooms. Florida’s State Board of Education adopted rules Sept. 16 that prohibit AI designed to simulate companionship or emotional relationships in K-12 schools and colleges. “Florida will not outsource childhood to a chatbot,” Gov. Ron DeSantis said. Source

THIS ISSUE

Trump says no. Some Meta AI calls were human.

Read the Reuters story

Brush Your Brain - The jingle

that started a movement

If you or someone you know is in crisis, call or text 988 (Suicide and Crisis Lifeline).

Jess Jessop is the Founder and CEO/CTO of Clinician Assist Inc. (BetterMind.Space), building the first voice-first AI-native mental health EHR with Casey Life and Peer AI Coach supervised by licensed therapists. A disabled veteran and 25-year AI/software engineering veteran, Jess brings lived experience as a mental health client to the mission of making daily mental health care as integrated as oral care.

ClinicianAssist.ai | BetterMind.Space | JessJessop.info | Subscribe | Archive

Trump Tells UN: No Global AI Control!

The news that moves policy, portfolios, and patient safety.

By Jess Jessop | September 23, 2026 | Issue #164

NO GLOBAL AI CONTROL, TRUMP TELLS UN.

Reply

Keep Reading

Conversational AI Watch