AI Psychosis in the C-Suite: When Tech Leaders Believe Their Own Hype
It started as a joke about chatbots. Now it's a diagnosis being applied — half-seriously — to the people running the world's largest technology companies. TechCrunch recently used the phrase AI psychosis to describe a particular condition spreading through Silicon Valley boardrooms: a kind of unshakeable, evangelistic belief that artificial intelligence is about to upend every job, every workflow and every business model, right now, immediately, and that anyone who hesitates will be left behind.
It would be easy to dismiss this as a turn of phrase. But the consequences are showing up in real decisions — mass layoffs, security shortcuts, public spats between billionaires about whether their software is conscious. For Australian workers and businesses watching this play out from a safe distance, it's worth understanding the psychology at work, because the fallout doesn't stay in California.
The diagnosis: belief mistaken for strategy
TechCrunch's framing of AI psychosis among CEOs isn't strictly clinical. It's shorthand for a pattern: leaders publicly insisting their workforce is on the verge of being automated away, that their products are about to achieve breakthroughs they can't quite demonstrate yet, and that any internal scepticism is a kind of heresy. The pitch deck has eaten the strategy memo.
That pattern matches what psychologists have long called groupthink — a phenomenon Irving Janis identified in the 1970s where cohesive in-groups make worse decisions because dissent is socially costly. In tech leadership today, the in-group is small, well-funded and constantly reinforcing one another at the same conferences, on the same podcasts, with the same investors. The story they tell each other becomes the story they tell shareholders, and eventually the story they tell their own staff right before the redundancy emails go out.
When the optimism hits the workforce
You can see the practical end of this in ClickUp's recent mass layoff, which TechCrunch held up as a signal about the future of work. The pattern is becoming familiar: a company announces deep cuts, frames them as AI-driven productivity gains, and tells investors that fewer humans plus more models equals a leaner, hotter business.
Sometimes that's true. Often, though, the AI productivity story is doing rhetorical work that the underlying technology can't yet support. Studies of enterprise AI deployments have repeatedly shown patchy returns, integration headaches and rising costs from running large models. But once a CEO has publicly committed to an AI-first transformation, reversing course feels like an admission of failure. So the layoffs proceed, the headcount is recast as a feature, and the workers absorb the cost of leadership's overconfidence.
For Australian companies — particularly the mid-sized firms that take their cues from US tech orthodoxy — there's a real temptation to import this playbook. It's worth pausing on. The Australian labour market is tighter, retraining is cheaper than rehiring, and the productivity dividends being promised in San Francisco haven't reliably arrived even there.
The security blind spot
The same overconfidence shows up in how companies handle the risks of the systems they're rushing to deploy. TechCrunch's reporting that everyone is navigating AI security in real time — even Google is a quietly damning admission. If the company with arguably the deepest machine-learning bench on earth is improvising its safety posture, what does that say about the thousands of smaller firms wiring large language models into customer service, internal search and code generation?
Prompt injection, data exfiltration through model outputs, hallucinated function calls — these are not edge cases. They are categories of vulnerability that didn't exist five years ago and for which the industry has no settled defences. A leader gripped by AI psychosis tends to treat these as engineering details to be solved later, after the launch, after the keynote. The opposite instinct — slowing down, threat-modelling, red-teaming — feels like cowardice in a culture that has decided speed is the only virtue.
'Seemingly conscious' AI and the limits of self-belief
The most striking sign that even insiders are getting nervous came from Microsoft's own AI chief. Mustafa Suleyman warned publicly about “seemingly conscious” AI — systems that aren't actually sentient but are persuasive enough that users, including sophisticated ones, start treating them as if they are. Suleyman's concern is partly about ordinary users forming unhealthy attachments to chatbots. But it cuts at executives too.
If a model can convince a lonely teenager it understands them, it can also convince a CEO that it's a strategic partner. Anyone who has watched a senior leader paste a sensitive document into ChatGPT and then quote its summary in a board meeting knows the dynamic. The fluency of these systems flatters the user. It tells you what you want to hear, in the register you want to hear it in. For someone already inclined to believe AI is about to remake civilisation, talking to one daily is not a reality check — it's a mirror.
What grounded leadership looks like
None of this is an argument against using AI. The technology is genuinely useful, and pretending otherwise is its own kind of denial. The argument is against the specific cognitive failure that comes when leadership stops being able to distinguish belief from evidence.
A few practical antidotes are visible in the better-run companies:
- Demand demonstrations, not narratives. If an AI initiative is going to justify cutting staff, it should be able to show measured productivity gains in production, not slides about projected ones.
- Protect internal dissent. The engineer who says “this model hallucinates on our customer data” is more valuable than the consultant who says “this is the biggest shift since the internet.” Groupthink dies when disagreement is cheap.
- Treat security as a precondition, not a phase. Google's improvisation should be a warning, not a template. Smaller organisations have less margin for error, not more.
- Notice when the AI is flattering you. Suleyman's “seemingly conscious” framing applies to executives too. A tool that always agrees with your strategy is not validating it.
The Australian angle
For readers in Australia, the temptation is to view all this as a uniquely Silicon Valley pathology — a problem for people with too much venture capital and not enough sleep. But Australian banks, telcos, retailers and government agencies are all making AI procurement decisions right now, often on the strength of pitches built inside the very culture TechCrunch is describing. The hype is being imported wholesale, sometimes by executives who've spent a week in San Francisco and come home convinced their org chart needs surgery.
The healthiest response isn't cynicism. It's calibration. AI will change knowledge work. It probably won't change it as fast, as cleanly, or as universally as the loudest voices claim. The leaders who'll look smart in five years are the ones treating the current moment as a long, messy integration project — not a religious awakening.
AI psychosis, if it's a real condition at all, is curable. The treatment is the same as it's always been for executive overconfidence: more data, more dissent, and a little humility about the gap between what a demo can do and what a business actually runs on.