Thiel-backed AI project to block bad press looks like a bust

There is a persistent belief in certain corners of the tech world that complex social problems can be eliminated through computation. Right now, there are plenty of aggrieved founders, venture capitalists and MAGA-adjacent influencers who feel that journalism is one of those messy problems. And so a group of powerful men in Silicon Valley with a demonstrated willingness to retaliate are trying, once again, to bend the Fourth Estate to their will.

The pitch is that better tools would enable us to separate fact from falsehood more efficiently. Give it enough data, the argument goes, and the machine will find a bias-free path to truth. It sounds like the kind of solution a public exhausted by misinformation and trained to scapegoat journalists might embrace: a technological fix for a very human problem. It also assumes that the people building these tools have no stake in the outcomes they produce.

The latest incarnation of this belief system arrives in the form of Objection AI, a project that presents itself as a kind of “truth tribunal” for journalism. The program is the brainchild of Aron D’Souza, an Australian lawyer whose most notable professional achievement remains his role in helping orchestrate entrepreneur and investment capitalist Peter Thiel’s legal strategy to secretly bankroll the lawsuit that destroyed Gawker. (After the outlet outed Thiel as gay in 2007, he backed former professional wrestler Hulk Hogan’s successful privacy lawsuit for publishing his sex tape.) The money behind Objection comes from that same ecosystem: investors like Thiel and Balaji Srinivasan, a crypto evangelist and prediction-market enthusiast.

Srinivasan, for his part, has been working toward something like this for years. Long before the current artificial intelligence boom, he was experimenting with the idea of turning truth into a market mechanism. In 2020, he proposed systems in which people could pay, using cryptocurrency, to “vote” on the validity of a claim, effectively turning factual disputes into speculative assets. Around the same time, Srinivasan was captivated by early large language models, predicting they would soon replace entire categories of journalism. Sports reporting could be generated from box scores, financial reporting from ticker data and movie reviews from captions. 

In this worldview, journalism is not a public service — it is a power center. And like any power center, it must be disrupted.

Objection AI’s logistics are startling. For a starting fee of $2,000, anyone can file a complaint against a piece of journalism, even if they are not the subject of the article. It can be a competitor, a political ally or a stranger with a random grievance. Once the complaint is filed, a team of investigators — described by D’Souza as including former FBI, CIA and National Security Agency officials — assembles an evidence file. The journalist is invited to defend their reporting. Then the material is handed over to what Objection calls its “AI tribunal”: a collection of large language models from major AI companies, coordinated by a proprietary system. The tribunal issues a so-called verdict on the factual claims in the story. That verdict feeds into a public score, the “Honor Index,” a numerical rating attached to the journalist’s name, and marketed as a measure of their integrity and track record. 

This is the same logic that made frivolous defamation suits useful tools of intimidation for decades. But, as D’Souza rather proudly observes, the adjudication process is stripped of the inconvenience of courts: “The Gawker litigation took ten years and millions of dollars. Objection industrializes this process.”

Speed is a large part of the appeal. Where traditional defamation cases require standing, evidence standards and years of expensive procedure, Objection promises a verdict in days. 

The ability to rapidly generate a counter-narrative, complete with the imprimatur of an “AI tribunal,” is tailor-made for an attention economy where perception often outruns verification. The PR win of a “vindicated” post on X with an Honor Index update arrives before the news cycle moves on, while the narrative can still be managed.

The incentives embedded in Objection’s system are clear. Documentary evidence is rewarded. On-the-record statements are privileged. Anything that cannot be neatly packaged into verifiable artifacts is treated with suspicion. D’Souza has suggested that a “scientific method” approach to journalism would simply eliminate anonymity altogether, arguing that if a source cannot be named, their information should not count. What this ignores is that some of the most consequential reporting in modern history — from the Watergate scandal that brought down Richard Nixon’s presidency to the initial reports of torture at Abu Ghraib — has relied on anonymous sources. 

The AI models assembled into Objection’s tribunal do not understand what an editor does when deciding whether to publish an anonymously-sourced story — how they must know the source’s identity and consider their potential motivations, carefully weigh the claims against known facts, find a separate independent source or documentation for corroboration, and then decide whether the circumstances merit publication. The “truth tribunal” does not know the difference between a source protected because they are telling the truth and a source who is anonymous because they are lying. These systems are — as has been extensively and embarrassingly documented — prone to generating fabricated citations, misreading evidentiary context and issuing confident-sounding verdicts about matters they fundamentally do not understand. 

Want more sharp takes on politics? Sign up for our free newsletter, Standing Room Only, written by Amanda Marcotte, now also a weekly show on YouTube or wherever you get your podcasts.

D’Souza has cited a University of Chicago study to argue that AI applies the law with perfect accuracy. But what the study actually found is that Generative Pre-trained Transformer models are more likely to rule based on judicial precedent, while human judges are more likely to be swayed by sympathetic contextual detail — a tension the study’s own authors declined to resolve in favor of either approach. 

Then there is the arbitration agreement — perhaps the most clarifying feature of the whole enterprise. If a journalist wants an on-the-record interview with someone in Objection’s orbit, they are asked to sign an agreement consenting in advance to the tribunal’s jurisdiction. They are asked, in other words, to preemptively accept the possibility of financial penalties assessed by an AI jury and overseen by Thiel, the man who destroyed Gawker, in exchange for a quote from someone who is plainly hoping to pick a fight. Or don’t sign, and risk being labeled uncooperative or unaccountable. 

It is a nonstarter.

No serious journalist is going to agree to a system that undermines their independence and subjects their work to the judgment of a private AI panel funded by individuals with a documented hostility toward the press. The idea that access to a source should come with conditions that compromise editorial integrity is fundamentally incompatible with how journalism operates.

So what happens when journalists refuse? Nothing, really. And that is the fatal flaw.

Without buy-in from the press, Objection AI becomes performative. Complaints can still be filed. Investigations can still be conducted. AI models can still generate verdicts. But those judgments have no binding force. They do not compel retractions. They do not impose damages. They do not alter the underlying reporting. The only output is a claim of vindication circulating in the same online ecosystems that already thrive on amplification and grievance. The point is the ask itself; the atmosphere it creates is an implicit signal that reporters who cover these men are already on notice. The potential chilling effect is the product.

This is why, despite the money and the branding and the rhetoric, the project looks like a bust. Not because the technology doesn’t function, but because the premise does not hold. You cannot build a parallel system of journalistic accountability and expect it to matter if the people you are trying to regulate simply opt out.

None of this is occurring in a vacuum. We are living through one of the most coordinated and well-resourced assaults on press freedom in American history. The Trump administration has defunded public broadcasters, excluded reporters from briefings and filed a $10 billion defamation suit against the Wall Street Journal. And it’s into this environment that Peter Thiel, who has made no secret of his contempt for democracy, is launching a system explicitly designed to make adversarial journalism more expensive and more hazardous. 

The real cost of Objection AI is not what it does. It is, as yet, a dud — a $2,000 slot machine that pays out a Community Note dressed as a verdict, in a process that compels nothing and binds no one. The real cost is what it normalizes. 

Every journalist who knows that a complaint against their story can generate a permanent public score attached to their name is a journalist who has been given one more reason to hesitate before reporting on a wealthy and powerful subject. The chatbot tribunal is, ultimately, a prop in a theater of intimidation — and the men behind it know exactly what they are producing. What they have not realized is that journalists have watched this performance before.