Jordan BlakeTechnology & Business Writer

Evening AI News Recap — Monday, March 30, 2026

Good evening. Today’s final digest closes out a Monday that has cut across the full spectrum of AI’s societal footprint — from courtrooms and research conferences this morning, to battery strategy and documentary criticism this afternoon. Tonight’s stories take a harder turn: algorithmic surveillance inside the federal government, a California biotech startup with plans that sound like science fiction but are deadly serious, and a long-running debate about lie detection that AI may be about to reignite. Buckle up.

What connects these three stories isn’t the technology — it’s the question underneath each of them: who decides when AI gets to make consequential judgments about human beings? A Palantir tool that surfaces IRS audit targets, a synthetic biology platform that might one day grow backup bodies, and AI-driven deception detection all sit at the edge of what democratic societies have agreed, so far, to permit. None of them have clear legal frameworks. All of them are moving forward anyway.

Image: AI-generated

1. Palantir Inside the IRS: AI That Decides Who Gets Audited

Documents obtained by Wired via public records request reveal that the Internal Revenue Service paid Palantir Technologies $1.8 million last year to improve a custom software tool designed to identify the “highest-value” cases for audits, unpaid-tax collection, and potential criminal investigations. The IRS contract was signed against a backdrop of institutional frustration: the agency runs more than 100 business systems and 700 data methods, most built over decades, creating a “fragmented landscape” that leads to duplicated effort, coverage gaps, and — in the agency’s own words — “suboptimal case selection.”

Palantir’s platform promises to cut through that fragmentation, aggregating signals from disparate legacy systems into a single ranked view of targets. The contract reportedly focuses on the clean-energy tax credits introduced under the Inflation Reduction Act — a politically charged area where the current administration has shown appetite for aggressive enforcement. Critics have raised immediate concerns: if the underlying data reflects historical audit disparities (Black taxpayers have been audited at significantly higher rates than white taxpayers at the same income levels), feeding that data into an optimization tool could systematize those disparities rather than correct them. A tool designed to find “highest-value” targets will optimize for whatever patterns are in the training data — and those patterns are not neutral.

The deeper issue is accountability. Palantir’s software is not a rubber stamp; it presents ranked cases to human agents who make final decisions. But in practice, algorithmic prioritization shapes decisions profoundly — agents presented with a ranked list tend to work down it. The question of how much weight the AI score carries in practice, versus how much independent judgment agents exercise, is exactly the kind of detail that FOIA requests rarely surface. Wired’s reporting opens the door; the harder investigation of how the tool actually operates in fieldwork remains undone.

Why it matters: The IRS-Palantir contract is a window into a broader pattern: government agencies are quietly building AI-driven case prioritization tools across enforcement contexts — tax, immigration, benefits fraud, child services. These tools make consequential decisions about which people get scrutinized, which get left alone. Deploying them without transparency, without auditable bias assessment, and without public debate is a governance failure in the making. The fact that WIRED had to use a FOIA request to surface even the basic contract details tells you everything about how these decisions are being made. Tools like OpenRouter are making multi-model AI pipelines accessible to smaller organizations — but the harder problem isn’t building these tools, it’s deciding who gets oversight of them.

📎 Read the full WIRED investigation: The IRS Wants Smarter Audits. Palantir Could Help Decide Who Gets Flagged

2. Brainless Clones and the Startup That Doesn’t Want You to Know About It

R3 Bio, a stealth biotech startup based in Richmond, California, surfaced briefly last week to announce it had raised funding to grow nonsentient monkey “organ sacks” as a cruelty-free alternative to animal testing. That was the approved story. MIT Technology Review has now reported what R3 didn’t want disclosed: its founder, John Schloendorn, has also pitched a far more radical vision — AI-assisted cultivation of “brainless human clones” as backup bodies for life extension, with the long-term goal of brain transplantation into a younger body as a route to a second lifespan.

The technical concept draws on a real birth defect — anencephaly — in which children are born without most of their cortical hemispheres. Schloendorn has reportedly shown medical scans of these children to potential backers as proof-of-concept: a body can live without much of a brain. His version would go further, engineering a biological substrate — essentially a body grown to biological maturity without higher brain function — that could serve as a replacement vessel for an aging mind, or a source of organs perfectly matched to the recipient’s genome. The startup has raised money from billionaire Tim Draper, Singapore-based Immortal Dragons, and life-extension fund LongGame Ventures.

AI enters this picture in multiple ways. Machine learning is central to the development of gene editing protocols, embryo development monitoring, and the complex protein engineering that would be needed to suppress cortical development while keeping the rest of the body viable. The more speculative brain-transplant scenario would require AI-guided surgical robotics that don’t yet exist. None of this is remotely near clinical reality — but the fact that it’s being funded, pitched to investors, and actively kept secret from the public speaks to how frontier longevity research is advancing in the shadows of the regulatory system, not under it.

Image: AI-generated

Why it matters: The R3 Bio story is a case study in how the most ethically fraught biotechnology research gets funded and advanced: quietly, among a small community of true believers, deliberately below the threshold of public scrutiny. The startup’s founders clearly understood that the brainless-clone pitch would generate backlash — which is why they suppressed it while pursuing the more palatable animal-testing narrative. But the suppression itself is the story. When research with profound ethical implications is being conducted in secrecy specifically to avoid the ethical debate, that is a governance problem that can’t be addressed retroactively, after the technology is mature. The regulatory frameworks for this kind of synthetic biology do not exist. Building them now — before the tools are ready — is vastly preferable to building them in a crisis.

📎 Read the full MIT Technology Review investigation: Inside the stealthy startup that pitched brainless human clones

3. Polygraphs Are Broken. Is AI the Fix — or Just a Better-Dressed Pseudoscience?

Polygraphs have been scientifically discredited for decades. Research consistently shows false-positive rates in the range of 20–30% — meaning one in four or five truthful people will be flagged as deceptive. George Maschke, who co-founded the advocacy site AntiPolygraph.org after a flawed polygraph ended his FBI application in 1995, is one of thousands of people whose careers were derailed by a technology the scientific consensus says doesn’t reliably work. And yet polygraphs remain in widespread use in US federal employment screening, law enforcement, and national security contexts. The legal and institutional momentum behind them has simply outrun the science.

Ars Technica’s investigation this week asks: given all that, is AI a better alternative? Researchers are exploring several candidates. Functional MRI-based deception detection looks for neural signatures of lying in brain activity data. Thermal imaging tracks micro-fluctuations in facial blood flow that may correlate with stress. Computational analysis of voice patterns, eye movements, and micro-expressions has also attracted research funding. Some of these approaches achieve better-than-chance results in controlled laboratory settings. But the same fundamental problem applies: in the real world, with real stakes and real anxiety, distinguishing deception from stress is extraordinarily difficult, and the physiological signals that indicate lying also indicate nervousness, which is what any innocent person feels when accused of lying in a formal interrogation.

The deeper epistemological problem — can we ever reliably detect deception from physiological signals? — has not been solved by AI, and may not be solvable in principle. What AI adds is statistical power and apparent scientific legitimacy. A neural network flagging someone as deceptive carries a kind of authority that a needle twitching on a paper chart doesn’t — even if the underlying validity is equally shaky. That authority gap is dangerous. Policymakers and courts that were appropriately skeptical of old-fashioned polygraphs may be less skeptical of AI-branded successors.

Why it matters: The AI lie detection story is a microcosm of a broader risk pattern: taking a discredited domain, adding machine learning, and repackaging the result as rigorous science. This pattern is appearing across high-stakes decision contexts — criminal sentencing risk scores, psychiatric diagnosis tools, social service eligibility algorithms. The question in each case is not whether the AI is more accurate than the baseline. It’s whether “more accurate than a broken tool” clears the bar for making consequential judgments about individual human beings. In most cases, it doesn’t — but the systems get deployed anyway because they offer plausible deniability and apparent objectivity. For teams building AI-assisted decision workflows, n8n provides the orchestration layer to build transparent, auditable pipelines — but transparency has to be a design goal, not an afterthought.

Analysis: The Accountability Deficit

Step back from the three stories tonight and a single thread emerges: algorithmic systems are being deployed to make consequential judgments about human beings — who gets audited, whose biology gets engineered, who gets flagged as a liar — in contexts where accountability structures don’t exist, don’t apply, or are being deliberately avoided. This is not a new observation, but tonight’s news illustrates how many different domains it now covers simultaneously.

The IRS-Palantir story is about algorithmic enforcement without auditable fairness standards. The R3 Bio story is about synthetic biology advancing under a veil of deliberate secrecy. The AI lie detection story is about the laundering of pseudoscience through machine-learning packaging. Each of these represents a different failure mode of the same underlying problem: consequential AI deployment in the absence of meaningful governance.

The response to this problem is not to ban the technologies. Palantir’s case prioritization could, in principle, be made fairer and more efficient than the current fragmented system — if it were subject to rigorous bias audits and public accountability. Synthetic biology research into organ cultivation has legitimate medical applications that could save many lives. Better deception-detection tools, if the science ever catches up, could serve justice. The problem is the absence of the “if” — the missing audits, the missing regulations, the missing public debate. That absence is a policy choice, and it’s one that is getting harder to undo the longer these systems are in the field.

What to Watch Tomorrow

• Congressional oversight: The Palantir-IRS contract was surfaced by FOIA, not by Congress — but expect it to draw attention on the Hill this week. Watch for statements from the House Ways and Means Committee or the Senate Finance Committee, which have jurisdiction over IRS operations.
• R3 Bio response: The MIT Technology Review exposé published today. R3 Bio and its investors have not publicly responded to the full scope of the reporting. A response — or continued silence — will tell its own story.
• DARPA expMath follow-up: The afternoon’s Axiom Math story touched on DARPA’s expMath initiative; watch for any forthcoming announcements on funded projects or benchmark results from that program, which is in active grant phase.
• UK AI Safety Institute: The international AI governance calendar has a cluster of EU and UK AI regulatory proceedings this week. Any formal guidance from the UK’s Safety Institute on high-risk AI applications in government enforcement contexts would land with particular relevance to the Palantir story.

That’s your evening recap for Monday, March 30. Three stories that should not leave you comfortable — which is exactly the point. Governance gaps in AI don’t announce themselves loudly. They surface in FOIA responses, investigative leaks, and scientific review papers that most people never read. Consider this your briefing. We’ll be back in the morning.

Image: AI-generated

This article was produced with the assistance of AI tools and reviewed by the AIStackDigest editorial team.