Adalma AI

AI agents for fraud, AML and sanctions. They assemble the case. Your officer signs it. Everything lands on an audit trail. Cernio · Cura · Sentra · Tessra

[Adalma AI @AdalmaAI]

Adalma builds AI for financial crime compliance.

The software runs in your environment. The agent assembles the case. A named officer signs it. Every filing lands on a tamper evident audit trail.

Cernio covers AML, BSA and FinCEN e-filing.

https://t.co/u8GmdOgNaD

[Adalma AI @AdalmaAI]

The full reconciliation, every figure with its source side by side: https://t.co/2FJx6o6lWl

[Adalma AI @AdalmaAI]

$38 billion lost to identity fraud and scams in 2025. That number is in an ad I got this morning, and it is accurate.

Here is what the ad leaves out.

[Adalma AI @AdalmaAI]

For scale: the FTC logged $15.9B for the same year. The FBI's IC3 logged $20.9B.

Different populations, same order of magnitude. Worth knowing before anyone quotes one number at you.

Source: Javelin Strategy & Research, 2026 Identity Fraud Study, 21 April 2026.

[Adalma AI @AdalmaAI]

Slowing the defence while the offence runs free isn't safety. It's a handicap.

In banking, slowness was never caution. It was a budget: proving what a system did is expensive, so change gets rationed.

Don't slow the technology. Make the evidence cheap.

[Adalma AI @AdalmaAI]

The proposal is to slow the frontier down. Review board, approval queue.

Fraud isn't waiting for one. Elder fraud, APP scams, synthetic identity, deepfaked voice on the verification call. All of it got better in 18 months. None of it went through a review board.

[Adalma AI @AdalmaAI]

Both sides are reasoning from first principles about what an approval regime does to challengers.

Financial services ran the experiment. A $4B credit union carries a big bank's BSA obligations with four analysts.

Doesn't stop them. Taxes them, in the one thing they can't buy.

[David Sacks @DavidSacks]

Some thoughts on Dario’s post:

1. Dario does not actually address Gavin Baker’s account of what he said – something he could easily deny if it were inaccurate.

2. Dario claims his critics live in a “bubble” where all regulation equals regulatory capture. He calls this an overly simplified view and notes that “Many people outside this bubble think of regulation as something that constrains corporate power and benefits ordinary people.” This argument is a straw man. Of course treating all regulation as capture would be overly simplified – but almost no one holds that view. I have repeatedly argued for strong antitrust enforcement to keep industries competitive, especially Big Tech. If Anthropic continues toward monopoly or duopoly status, I would be among the first to demand those rules apply.

3. Regulatory capture is not vague or in the eye of the beholder. Nobel laureate George Stigler defined it as regulation acquired by an industry and designed and operated primarily for its benefit. Stigler challenged the traditional view that government regulation arises from a benevolent state protecting the public from market failures. Rather, industry groups have concentrated stakes and pour resources into influencing regulators, whereas the public’s stake is diffuse and unorganized. The revolving door between companies and the agencies that regulate them compounds the problem. Anthropic understands these dynamics: it has hired multiple senior Biden AI-policy officials and built a substantial government-affairs operation plus a network of aligned organizations to push its preferred frameworks at state and federal levels.

4. Dario has consistently pushed for a new federal agency to review and approve frontier models prior to release – a proposal framed variously as an “FDA for AI,” an “FAA for AI,” and most recently a “FINRA for AI.” I call it a “DMV for AI” because a review process modeled on the FAA or FDA (which takes years) or FINRA (which issues rules for a staid industry widely seen as protecting incumbents) will create long queues as AI models wait for testing and approval. This process will only become more labyrinthine as rules accumulate to prevent theoretical harms. This would handicap the U.S. relative to China, which will not adopt the same constraints. It would also undermine Anthropic’s own business model, whose pricing power depends on remaining ahead of open models. Whatever Dario states today, it is difficult to believe the company would simply accept outcomes that erase that advantage.

5. Anthropic is on track to become one of the most valuable companies in history, with the resources to navigate any approval process and shape the rules while competitors wait. Dario wants open models under heavier scrutiny – he has called them dangerous in Senate testimony, criticized them for not being centrally monitored or withdrawn, and linked them to IP theft. He says he has never sought a ban, but he could achieve a similar result by insisting that identical rules apply to both open and closed models. The U.S. risks becoming an island of costly closed models while the rest of the world races ahead with broader choice.

6. Dario acknowledges that AI is structurally centralizing but attributes this mainly to chips and scaling laws. Access to compute matters, but the deeper risk is who decides which capabilities are available to whom. His preferred pre-deployment testing and FAA/FINRA-style oversight would place that gatekeeping power in a federal bureaucracy working hand-in-glove with a small number of frontier labs – reinforcing centralization rather than countering it.

7. The second part of Dario’s post assumes we have amnesia about Anthropic’s well-orchestrated campaigns hyping AI fears. His May 2025 claim that AI would wipe out 50 percent of entry-level knowledge jobs within five years still lacks supporting evidence fifteen months later. Similarly Anthropic breathlessly promoted its heavily contrived “blackmail” study on 60 Minutes. Yet Dario blames public negativity on a long-standing loss of trust in institutions rather than his own messaging.

8. These narratives have done more than anything to shape public fear. People are left asking the same question Mark Zuckerberg posed: why race to build a future you describe in such negative terms? Thomas Sowell’s "The Vision of the Anointed" captures the mindset – elite intellectuals convinced that only they are enlightened enough to control the outcome. As Zuckerberg notes, concentrating power in the hands of an enlightened few has rarely produced the promised results; the practitioners turn out to be less enlightened in practice than in self-conception.

9. Gavin Baker summarized the disagreement cleanly on our pod: Dario believes frontier AI is too powerful to distribute; we believe it is too powerful to centralize. Dario appears to believe, sincerely, that safety and progress are best served by centralizing authority in a marriage of corporate and state power. The weight of human history gives us reason to fear that outcome.

[Adalma AI @AdalmaAI]

We signed our first customer on exactly that work this month. Not a pilot. Production since 1 August.

If you run risk, compliance or engineering at a financial institution and want to see what it looks like in practice, our DMs are open.

[Adalma AI @AdalmaAI]

The order arrives as three questions. What did the model see. Which policy was in force at the time. Who signed their name to it.

Nobody answers those from a discovery call. You answer them inside the institution, on real data flows.

[Adalma AI @AdalmaAI]

A $4B credit union carries JPMorgan's exam, its sanctions obligations and its SAR volume with four analysts instead of four hundred.

Open weights hand you a model. Not a filing. Not an audit trail. Not a core integration.

That gap is the whole business.

[Adalma AI @AdalmaAI]

Two very sharp people spent the weekend arguing about the frontier.

Both were looking up. It is worth looking down.

[Adalma AI @AdalmaAI]

"intelligence too cheap to meter"

Wonderful. My compliance officer's questions are not getting 10x cheaper.

What data did that model see?
Which policy version was live?
Who approved it?
Show me the record. It is 14 months later.

The bill does not disappear. It moves to proof.

[Sholto Douglas @_sholtodouglas]

>may be really civilizationally useful in giving us time to adapt to what might be a fast takeoff.
Agree - this has really interesting implications on RSI, which is that even if you had Noam Shazeer level models tomorrow they'd still be experiment bottlenecked, so it changes the shape of takeoff. very important factor to model.
>But my argument would be the same: rather than worrying about inequality we should focus on growing the pie so fast that even the bottom of the income/compute distribution are much better off here in America than almost anywhere else.
1000% agree. My median expectation is that compute spend follows the 2-3x yoy trendline (so that next year it is >2T), and that a given level of intelligence keeps getting 10x cheaper every year due to efficiency improvements throughout the stack https://t.co/Xyz4KDqI14, which will hopefully compound into intelligence too cheap to meter.

Will take you up on that dinner in Boston sometime, or let me know next time you're in SF!
And don't worry - I drink, can't take the Australian out of the guy

[Adalma AI @AdalmaAI]

"every financial institution and critical infrastructure provider trying to white hat hack themselves"

Not a forecast. A two-year work order, landing on institutions that never asked for it. The bank with four analysts gets the same order as the bank with four hundred.

[Sholto Douglas @_sholtodouglas]

Awesome response - thank you for coming back with such good faith.

So much I agree with here. Also a lot of things I think the picture is more nuanced and I want to justify our takes more

A few quick ones, but I might post more here later.

I think that cyber is ultimately defence dominant, and if we put in the work over the next 2 years (tldr every financial institution and critical infrastructure provider trying to white hat hack themselves), then I have 0 concern for any open coding model. In the interim, a totally reasonable choice is ‘we’ll take substantially increased cyber attack risk for no imposition on our freedom’ - but the USG should be in a position to actively make that choice having measured and evaluated the risks for themselves.

Bio is offence dominant, and fixing that will probably take well into the 2030s - this means that we’re stuck with genuinely huge risks there which society needs to decide whether it wants to accept (I.e. world leading virologist in your pocket). It’s entirely possible that society says ‘yes those risks are worth it’, but I think society/the gov should have an arm stood up to take that seriously (e.g. run uplift trials where they see if AI helps them more than YouTube access, genuinely test what the worst thing someone could do with 200k and a garage is etc), evaluate it for themselves and make that call with each new capability level.

The second is why has Dario/us talked about risk so much? Fundamentally it’s because we’ve wanted to be honest with people. Employment risk is the classic here. I actually disagree with Dario on the pace - I think it’s most likely that compute shortages, diffusion complexity, policy and unmet demand for services mean that even for years after we have models which could automate 95% of computer facing jobs (models will get there in 28), people will work at them (well into 2030s) - but I do think we as a society should take the possibility far more seriously than we are now, and prepare contingency policies for what to do at various levels of unemployment (e.g. you could imagine not letting profitable companies lay off more than 5% per year), as well as METR style evals to measure progress on different job families so we have a clear picture. Our opinion has always been that we need to be straight up and honest with people.

Completely agree that as both a company and an industry we have utterly failed to present a positive picture of a future people want to fight for, and that this is actively decreasing our chance of getting to that future. So much we need to do better there.

One final half baked one - I don’t think the Mark essay engages with the actually hard part of the problem, which is that inequality of compute will matter far more than open models in having personal superintelligence. Will talk more about this later, but TLDR think the essay presents open models as a kind of panacea without getting into the heart of the issues I actually expect in the future

[Adalma AI @AdalmaAI]

@nikesharora asks who is responsible before we regulate AI.

Financial services answered already. A SAR is signed by a named officer at a named institution. No filing has ever been accepted from a model.

Cernio builds around the signature. The human attests, then it transmits.

[Adalma AI @AdalmaAI]

@nikesharora on open weight models from any country: "You would test them all"

You would, if you could. Testing means running member data through four vendors. Four disclosures, zero decisions.

Tessra brings models to your eval set. Test them all on your side of the boundary.

[Adalma AI @AdalmaAI]

@nikesharora: "One can't regulate for success."

In financial crime the rules already exist. BSA, OFAC and SR 11-7 predate AI and all apply to it.

The question is not whether AI gets regulated. It is whether your AI can produce the evidence today's rules demand.

That is Cernio.

[Adalma AI @AdalmaAI]

@nikesharora: "We have gone from full stack LLMs to wanting to separate models and context, evals,et al."

Right split. But hand both halves to one vendor and you only renamed the dependency.

Tessra keeps traces and evals inside your boundary. Swap the model, keep the context.