What this is about:
A new Wired report says OpenAI backed an Illinois bill that would sharply limit when frontier AI companies can be sued for catastrophic harm caused by their models. The bill, SB 3444, would shield developers from liability for “critical harms” — including mass death or at least $1 billion in property damage — so long as they did not act intentionally or recklessly and published a safety protocol and transparency report.

Why it matters:
This is a fairly stark regulatory ask: if your model helps cause enormous harm, publishing your own paperwork may be enough to narrow legal exposure. Critics quoted by Wired call it a “get-out-of-jail-free card.” Even Anthropic, hardly a natural home for anti-industry rhetoric, has reportedly opposed the bill.

What’s happening:
SB 3444 applies to “frontier” models trained above a compute threshold or at a cost above $100 million. It says developers are not liable for critical harms if they publish a safety and security protocol, publish a transparency report, and did not intentionally or recklessly cause the harm. It also lets firms count as compliant if they agree to relevant EU rules or enter a qualifying agreement with a federal agency.

The catch:
That is not quite the same thing as accountability. NIST’s AI Risk Management Framework is explicit that all involved AI actors share responsibilities for designing, developing, and deploying trustworthy systems. The point is not that developers own every downstream misuse, but that responsibility does not vanish the moment a model leaves the lab.

The bigger picture:
There is a serious policy debate to be had here. Frontier-model harms are hard to attribute cleanly, especially once tools are integrated into messy real-world systems. But the stronger academic proposals in this area tend to argue for more transparency, third-party validation, incident reporting, and clearer liability pathways for catastrophic risk — not less. Recent work on catastrophic AI liability makes that case directly, drawing analogies with aviation, healthcare, and nuclear governance. The OECD is also pushing toward more standardised AI incident reporting, precisely because weak reporting makes accountability harder.

Reality check:
OpenAI’s position, as described by Wired, is that the bill is a pragmatic attempt to encourage labs to publish safety frameworks and avoid a patchwork of unworkable state rules. That is not a frivolous concern. But there is a large difference between avoiding regulatory chaos and pre-emptively softening liability for catastrophic harm.

The interesting bit:
The familiar AI line is that these systems may be powerful enough to transform society. The less glamorous follow-up is that companies seem rather less enthusiastic when that power is paired with legal responsibility.

The line:
If an AI company can help write the safety rules and then use those same rules to dodge the consequences, that is not governance. It is brand management with legislative support.

Sources: Wired | Illinois SB 3444 text | Illinois General Assembly bill page | NIST AI RMF 1.0 | OECD incident reporting framework | Catastrophic Liability: Managing Systemic Risks in Frontier AI Development

My amygdala is hijacked. For the uninitiated, that’s the bit of your brain that’s supposed to keep you alive deciding, quite unhelpfully, that everything is a threat. It’s a smoke alarm that’s gone off because you made toast. Overwhelm is not theoretical this week, it is fully operational.

I am now dreaming about the campaign trail. Not in a visionary, “this is my calling” way. In a “being chased down the road by opposition candidates” way. Which feels… about right. The brain does love a metaphor, and apparently mine has decided subtlety is no longer required.

We also finished Everwood. Which means my designated happy place has now been formally extinguished. My husband, clearly recognising the level of situation we are dealing with, has gently offered that we can just… start it again and carry on watching it… for as long as I want… from the beginning (He’s a keeper, no really, he is). I mean. Sh*tballs. This is not a normal sentence to be said in a well-regulated household. But also, I am not ruling it out.

And then there’s Finch. What started as a very professional, “this app has been recommended to some of my patients, I should understand what I’m dealing with” has rapidly devolved into a full dopamine chase to keep my Bird alive. I am now emotionally invested in the well-being of a pixelated creature who congratulates me for brushing my teeth. I don’t even know what to say about that.

I think what I’m circling is this: there is the version of me who holds complexity, thinks systemically, keeps showing up, keeps going. And then there is the version of me who needs a cartoon bird, a rewatch of Everwood, and to not be chased through suburban streets by people in blazers. Both appear to be present. Both, annoyingly, are real.

On the plus side, tonight we are starting season 2 of The Shrinking, for those of you who have followed along from the beginnings of our newsletter: I hope to be fully engaged back in my best ‘Paul’ mode soon!

Do you know of or have a great resource for therapists that we should highlight? Please email Ann at [email protected].

The tool: Prescribing time in nature.

Why it matters: According to Axios, social prescribing is on the rise: healthcare providers are increasingly referring people to non-pharmacy supports like walking clubs, art studios, choirs, gardening, lakeside activities, and other community-based programs.

The idea is not that nature or creative activities replace therapy or medication. It is that health is also shaped by connection, movement, sunlight, routine, community, and the parts of life that happen outside clinics.

How to use it: Offer a nature prescription as a small, specific experiment.

Make it concrete: Collaborate with the client to choose something realistic, accessible, and motivating

A few examples:

Research shows that even looking at nature images or watching nature videos has also been associated with improved mood and reduced stress, so this can be a lower-barrier option for clients with safety, mobility, weather, transportation, or access constraints.

Good for clients who are: Isolated, depressed, anxious, burned out, disconnected from their body, perfectionistic, stuck in rumination, or needing behavioral activation that does not sound like a productivity assignment.

Keep in mind: Not everyone has safe, nearby, beautiful, or disability-accessible green space. Nature prescriptions can include a sunny window, courtyard, library garden, weather, birdsong, sky, or even an audio recording of rain.

The finding: A highly supportive chatbot did not reduce loneliness as well as texting with a randomly selected real human peer.

What happened: A preregistered study in the Journal of Experimental Social Psychology assigned 296 first-semester university students to one of three groups for two weeks:

What they found: Only the students who texted with another person reported lower loneliness after two weeks. The chatbot helped students feel less negative in the moment, but that mood boost did not translate into reduced loneliness over time.

Why it matters: The chatbot was designed to be consistently warm and supportive, using principles from relationship science. And still, interactions with a random human peer offered something the chatbot could not.

The possible ingredient: Mutuality. The researchers suggest that human connection may reduce loneliness not only because we receive care, but because we get to offer something back. A chatbot can validate you. A person can need you, respond unpredictably, and make the exchange feel reciprocal.

Therapist takeaways:

Do you hear that?: It’s the collective “well, obviously” from the chorus of therapists who never doubted the power of a human relationship.

Source: Journal of Experimental Social Psychology, March 2026

This week, I’m off the clock cheering on my 9-year-old’s baseball team: the unfortunately, accidentally, and now permanently named Blue Ballers.

How it happened: The kids chose “Ballers” as their team name, then received their uniforms, which are, yes, blue. This led to a very earnest dugout full of young boys cheering, “Gooooo Blue Ballers!” and the dads/coaches trying to convince them to go with just “the Ballers,” which isn't sticking.

The kids have no idea about the double meaning, and this is perhaps one of my favorite moments of parenting ever.

Ann is writing Marianne’s Off the Clock for her this week because Marianne has been hustling her a— off every weekend for the past few months as she runs for re-election as Town Councillor.

She’s been knocking on doors, canvassing, holding meeting after meeting, and no doubt doing what Marianne does: kicking ass, asking thoughtful questions, and connecting with constituents like the deeply relational, systems-thinking human she is.

All of this is happening on top of her already intense job in the NHS. So this ‘Off the Clock’ is less about what Marianne is reading, watching, baking, or buying, and more about cheering her on.

Please send Marianne all the positive thoughts for the election on May 7th and the weeks leading up to it.

May the doors be friendly, the weather cooperative, her shoes supportive, and the voters wise.

Stories from the community

Here’s how things landed:

The spread reaffirms something we already knew: there are many ways this job is hard.

And the drain is not coming from just one place. It is administrative, emotional, relational, political, and structural, which might be why advice to “just practice self-care” tends to land flat.

Notes from the Community:

From someone who chose “Current events affecting clients and corresponding sense of helpelessness. “ They wrote:

“I'm so tired of living through the hellscape. It's hard to ever get a break from it all!”