Care in the Machine No. 7: The People Who Built It Are Asking Themselves to Slow Down

Care in the Machine is Dr. Michael Jones's weekly note on counseling, ethics, and the technology showing up in our work. It arrives by email every Monday. Subscribe here, or browse past issues.

Care in the Machine
A weekly note on counseling, ethics, and the technology showing up in our work.
Issue No. 7 · Monday, September 14, 2026

On September 12, Anthropic CEO Dario Amodei published an essay asking the rest of the AI industry to slow down. He called it pacing the frontier: keep building, but let the safety work catch up to the capability first. Sam Altman said OpenAI needs to do the same thing. Elon Musk, who competes with both of them, posted three words: Dario is right.

Three people who are fighting for the same customers stood up in public this week and agreed that the pace itself is the problem. Amodei went further and named a specific fear, that a swarm of rogue AI agents could do real damage across the internet within the next six to twelve months if nobody paces this correctly.

This newsletter has spent this whole season making one argument about pace: the care ethics model I keep returning to here says adoption should answer to the relationship, not to the release calendar. This week that argument found some unlikely company. Anthropic's own CEO just told his industry to slow down. Pacing ourselves in counseling was never about falling behind. It was about doing this correctly.

This week's teaching ideas

For the therapist. Give yourself a plain rule before you adopt a new AI tool in your practice: not on launch week. Wait until someone outside the company that built it has actually looked at what it does with data and how it behaves under pressure. Amodei himself just told his own industry he cannot rule out serious harm in the next year. A vendor's demo is not more evidence than that.

For the supervisor. Ask your supervisees what AI tools they have started using in the last month, and how they decided to trust them. If the honest answer is it seemed helpful and I was busy, that is not a decision, that is a shortcut. Gatekeeping right now means teaching people to pace themselves the way the industry itself just said it should.

For the counselor educator. Teach the difference between capable and ready. A tool can do something impressive in a demonstration and still not be ready for a room with a real client in it. Use this week's essay as the case study directly: if the CEO of Anthropic will not vouch for his own industry's pace, a student should not assume a vendor's pace is safe for theirs.

For the counselor in training. Before you use any AI tool with client information, even something administrative, ask your supervisor two questions out loud. What does this tool do with the data. Has anyone outside the company reviewed it. If you cannot answer both, you are not ready to use it yet, and that is not a failure. That is pacing yourself correctly.

This week's free resource: A Pacing Checklist

This one comes with the receipts. One page: a short, sourced rundown of what Amodei's essay actually said and who backed it, Anthropic's evaluator-access commitment, Altman's agreement, Musk's three words, then six questions to run before you adopt any AI tool into your practice, your teaching, or your admin work. Print it, hand it to a supervisee, or use it to open the conversation in class the next time this week's news comes up.

Download it here (no form, no gate)

Subscribers get the resource free every week. That is the deal: you signed up, you get the tools.

From my desk

The last couple of weeks I have been using a connector to hold the moving parts of a research proposal I am building with colleagues at a couple of other institutions, on measuring wellbeing. Not a course, not a training. An actual grant application, with a grants-office question here, an IRB question there, and more than one email thread about who is actually based where these days.

I asked the assistant to pull the last several emails on the project and build me a table: task, owner, date. It did that part fine.

Here is where I kept my hands on the wheel, and where this connects to everything above. The table looked clean, and clean is exactly when I get suspicious. Who is really the co-PI at which institution, what the IRB actually said, those are not things I take from a summary. I checked every row against the actual emails before I acted on a single one of them. Pacing myself did not mean refusing the tool. It meant not letting a clean-looking output stand in for verification.

Try this: pick one project with a scattered email trail. Ask your assistant to build a task, owner, date table from the last five messages on it. Then read every row against the source before you trust it. The table saves you time. The checking is still yours.

This week in the field

Patients notice when a tool is in the room, even when you tell them, and the noticing does not go away just because your practice adopted the tool quickly. Sidney Shaw and Darin Bell asked people to imagine hypothetical therapy scenarios and rate how acceptable each one felt. Human-only therapy scored highest, 6.57 on a seven-point scale. Fully disclosed AI use in a text-based or telehealth scenario dropped that by 2.89 and 3.09 points. Undisclosed AI use dropped it further still, by 4.65 and 4.80 points. The authors conclude that full disclosure does not restore a client's sense of authentic engagement (Shaw & Bell, 2026). So what for us: speed of adoption and quality of consent are not the same project. You can move fast on the first and still fail the second.

California's SB 903 is sitting on Governor Newsom's desk as of this writing, enrolled and sent to him September 9, with a September 30 deadline to sign or veto. The bill would keep AI systems from independently marketing or providing psychotherapy and would require disclosure and consent before AI records or transcribes a session. It is, in its own way, a pacing document, a legislature trying to catch up to a pace the market already set. Worth watching whether it becomes law before the industry's own pacing conversation cools off.

And a dated one worth having on the shelf if you supervise: a review published back in June maps existing supervision guidance onto AI-assisted tools, chatbots, pattern detection, session tracking, and lands on the supervisory working alliance as the thing most at risk (Brinck et al., 2026). It is not new, it published eleven weeks ago and only resurfaced in this week's searches, so I want to be straight about that. But it is a real synthesis, and if you are building any kind of policy for AI-assisted supervision at your site, it is a reasonable place to start reading.

An idea to float

Here is what I am turning over. Three competing AI companies just wrote something like a pacing statement for themselves, in public, under real competitive pressure not to. I do not know of a counseling organization that has done the same thing for our own adoption of these tools. Not a position paper about AI in the abstract, an actual pacing statement: here is what we will not adopt until it clears this bar, here is who reviews it first, here is how fast we are willing to move. I am not sure our field is organized enough yet to write one. I am floating it here because I think it is worth finding out.

Stay curious,
Dr. Jones
Michael Jones, PhD, LPC-S, NCC, BC-TMH

Renewing your BC-TMH?

Three short courses at the Counselor Education Collective count toward renewal, each two clock hours, each $39:

• Practicing Across State Lines: Law, Ethics, and Safety in Telemental Health
• AI at the Screen: Ethical Use of AI Tools in Teletherapy
• Culturally Responsive Care: Presence Across the Screen

Continuing education is provided by Renewed Vision Counseling Services, NBCC-Approved Continuing Education Provider, ACEP No. 6549. The Counselor Education Collective is the storefront for all courses.

Supervising through a screen? The Telemental Health Supervision Certificate is live: six hours on alliance, ethics, documentation, and gatekeeping at a distance.

Save the date: PESI's AI in Clinical Practice conference, October 29 and 30, 2026. I will be there; registration and details here.

More of my work at PESI. My trainings there run from AI and the ethics of digital afterlives, with Megan Devine, to advanced grief counseling. Browse my PESI catalogue

References

Brinck, E. A., Soldner, J. L., Kuo, H. J., Sabella, S. A., Landon, T. J., Bernacchio, C. P., & Boland, E. A. (2026). An AI perspective on counseling supervision. Behavioral Sciences, 16(6), Article 1038. https://doi.org/10.3390/bs16061038

Shaw, S., & Bell, D. (2026). Acceptability of AI-assisted psychotherapy to prospective patients and their intent to continue treatment: A vignette study. American Journal of Psychotherapy. Advance online publication. https://doi.org/10.1176/appi.psychotherapy.20260007

News items on Amodei's essay and the industry response (September 12, 2026) are drawn from reporting by Axios, CNN Business, the Washington Post, Bloomberg, and CNBC, per this newsletter's standard practice for policy and industry news.

Care in the Machine arrives by email every Monday, with teaching ideas, field notes, and a free resource each week. Subscribe here · Browse past issues