The Statements You've Forgotten That AI Hasn't

2026-05-09

In 2023, a mid-sized fintech published a press release describing its platform as "powered by proprietary AI that eliminates human error in credit decisions." It was the language of the moment — credible, forward-leaning, carefully approved by legal. Two years later, the company had quietly repositioned. The AI element was still there, but the claims had been moderated in line with what the technology could actually do.

The press release, however, had not been quietly repositioned. It was still being indexed. It was still being cited. And when an institutional investor used a public AI system to research the company's technology stack before a Series C call, the AI returned the 2023 language — confident, unqualified, presented as current — as its primary characterisation.

The investor arrived at the call with questions the company no longer had the same answers to.

This is one of the least-discussed risks in corporate AI governance: the way that historical communications become permanent inputs into public AI systems, and how those systems then surface that content without temporal context, promotional framing, or the disclaimer that sat at the bottom of the original document.

Forward-looking statements come with legal wrappers in regulated filings. They do not come with those wrappers when a public AI system paraphrases them in response to a due diligence query.

Earnings call transcripts, investor presentations, press releases, LinkedIn posts from senior leaders — all of it is potential source material for the systems that stakeholders now query as their first research step. The content enters these systems as communications designed to persuade. It exits as apparently neutral fact.

The new disclosure risk nobody briefed IR on

The past 12 months have seen a significant rise in AI-related securities litigation. Harvard Law's Corporate Governance forum identified "agent washing" — inflated claims about AI agent capabilities — as an emerging securities disclosure risk in April 2026. The SEC has been active on AI-washing enforcement since 2024. Regulators are increasingly able to cross-reference what a company said in its disclosures against what public AI systems serve to investors when those investors come asking questions.

The exposure runs in both directions. A company that overstated its AI capabilities faces enforcement risk from regulators who notice the gap. It also faces a subtler, ongoing risk: those overstatements are still circulating in the public AI systems that investors, journalists, and counterparties use to build their first picture of the organisation. The original claim may have been retired in official communications. In AI-generated research, it has not.

Investor relations teams brief carefully for earnings calls. They prepare holding statements for difficult questions. Very few have mapped what a public AI system currently says about their organisation's capabilities — or how old the source material is.

Context is the first casualty

When public AI systems generate a summary of a company's capabilities, they typically do not distinguish between a 2024 earnings call and a 2019 product announcement, between a CEO interview and a speculative analyst note, between language that was accurate when published and language that was always aspirational.

What the investor sees is a synthesis. What drives that synthesis is frequency and apparent authority — not editorial judgment about what the organisation currently wants to communicate.

The practical implication: organisations that have undergone significant transformation, repositioning, or technology evolution since their last major communications push may be systematically misrepresented in AI-generated research. Not because of hallucination — but because of accuracy. What the AI says may have been true once. The problem is that "once" is invisible.

A repositioned organisation and an organisation that overstated its capabilities look identical inside a public AI system. Both are being described by their own past words, served without the context that would make those words interpretable.

Clarity starts with visibility

Addressing this risk doesn't begin with a communications campaign or a legal review of past filings. It begins with knowing what public AI systems currently say — across investor, regulatory, and media lenses — and mapping that against what the organisation wants those audiences to understand today.

Perception in AI systems is not fixed. It shifts as new information enters the ecosystem. But it cannot be shifted without first being seen. Organisations that don't monitor what public AI systems are saying about them have no way to identify the gap — or assess how wide it has grown since the last investor roadshow.

The press release from 2023 is still out there. The question is whether it still represents you — and whether anyone has thought to check.