Deep Dive

Consumer AI Is Over: Why Big Labs Are Abandoning Users for Government Contracts

OpenAI kills GPT-4o for consumers, shelves adult mode, chases DoD deals. Anthropic courts banks while flagged as a security risk. The pivot is real.

Gradient team

— 7 min read
Consumer AI Is Over: Why Big Labs Are Abandoning Users for Government Contracts
Photo by Vitaly Gariev on Unsplash

Editorial digest April 13, 2026
Last updated : 00:16

TL;DR

  • OpenAI has quietly retired GPT-4o from ChatGPT while keeping it on the API for enterprise clients — the clearest signal yet that consumer users are subsidized marketing, not the business model
  • Anthropic is simultaneously courted by Trump officials for banking AI and flagged by the DoD as a supply-chain risk, exposing the incoherence of Washington's AI strategy
  • Tech valuations have returned to pre-AI boom levels according to Apollo data, suggesting the market is repricing AI from "paradigm shift" to "useful tool" — a healthier but painful adjustment
  • OpenAI's abandoned "adult mode" promise isn't about content modesty — it's about not doing anything that complicates government and enterprise sales cycles
  • India's frugal AI models offer an alternative path: smaller, cheaper, task-specific — and increasingly what enterprise buyers actually want

Why it matters now

Three things converged this week that, taken individually, look like noise. Together, they're a pattern.

OpenAI pulled GPT-4o from ChatGPT's consumer interface while keeping it alive on the API — the product enterprise customers pay for. Sam Altman's much-hyped "adult mode" has quietly disappeared from the roadmap, coinciding with OpenAI's deepening government relationships. Meanwhile, TechCrunch reports that Trump administration officials are actively encouraging banks to pilot Anthropic's Mythos model — the same Anthropic that the Department of Defense recently labeled a supply-chain risk.

And the backdrop: Apollo's data shows tech valuations have deflated back to pre-AI boom levels. The froth is gone. What remains is a sector recalibrating around a simple question: who actually pays for AI?

The answer, increasingly, is not you and me. It's the DoD, JPMorgan, and Netflix's content recommendation engine. The consumer AI era — the one where ChatGPT was going to be everyone's personal assistant — is being quietly unwound in favor of something more familiar: enterprise software sales with government contracts as the anchor tenant.

What actually happened

The GPT-4o disappearance. OpenAI removed GPT-4o as a selectable model from ChatGPT's consumer product. The model remains fully available through the API, where enterprise and developer customers access it. Reddit users noticed the change, with some confusion about whether the model was fully deprecated. It isn't — it's just no longer for you, the free or Plus subscriber. OpenAI has not issued a public explanation beyond the standard model-rotation framing.

The adult mode that never shipped. Months ago, Sam Altman publicly stated he wanted to "treat adults like adults" — an unlockable mode with relaxed content filters for verified users. As one Reddit user catalogued, not only has this feature never materialized, content filters have gotten progressively stricter. The timing correlates with OpenAI's DoD engagement and its stated pivot toward business customers. The Sora 2 consumer product was also shelved around the same period.

Anthropic's contradictory Washington moment. Per TechCrunch, Trump administration officials have been encouraging major banks to test Anthropic's Mythos model for financial applications. This is remarkable given that the Department of Defense recently declared Anthropic a supply-chain risk — a designation that typically chills, not encourages, government-adjacent adoption. No public reconciliation of these positions has been offered.

The valuation reset. Apollo's analysis shows tech sector valuations have returned to levels seen before the AI investment surge. This isn't a crash — it's a normalization. The "AI will eat everything" premium has been priced out. What remains are companies being valued on revenue trajectories and margins, not vibes and demo videos.

Strategic analysis

The enterprise pivot is structural, not tactical

OpenAI's consumer moves make perfect sense if you stop thinking of ChatGPT as a consumer product and start thinking of it as a lead-generation funnel. Free and Plus users demonstrate demand. API customers generate revenue. When these two goals conflict — as they do with content moderation, model availability, and feature prioritization — revenue wins.

The "adult mode" abandonment is instructive. The feature would have been popular with consumers. It would have been catastrophic for enterprise sales. Imagine pitching ChatGPT integration to a Fortune 500 compliance department while simultaneously running an "unfiltered mode" for individual users. The sales conversation writes itself — badly.

This is the classic platform lifecycle: attract users with a generous consumer product, then redirect engineering toward the customers who pay enterprise margins. Microsoft did it with Office. Slack did it. Now OpenAI is doing it, just faster.

Washington's AI policy is running on two operating systems

The Anthropic situation reveals something important about how Washington currently processes AI. Different parts of the government are making contradictory decisions because there is no unified AI procurement or risk framework that actually binds across agencies.

The DoD flags Anthropic as a supply-chain risk — likely related to foreign investment concerns or data-handling practices. Meanwhile, Treasury-adjacent officials are pushing the same company's models into the banking system. Both decisions may be individually rational. Together, they signal that AI vendors can shop for friendly regulators within the same government, and that enterprise customers face genuine confusion about which signal to follow.

For Anthropic, this is both opportunity and risk. Being simultaneously courted and flagged creates a precarious position where any policy consolidation could break in either direction.

What the valuation reset actually means

The Apollo data deserves careful reading. Tech valuations returning to pre-boom levels doesn't mean AI failed. It means the market has moved from pricing AI as a revolution to pricing it as an input — a productivity tool that improves margins incrementally rather than creating entirely new markets overnight.

This is healthier. It's also painful for anyone who raised money at 2024-2025 valuations. The gap between "AI as transformative technology" (true, over decades) and "AI as immediate revenue multiplier" (mostly not true yet) is where capital gets destroyed.

India's frugal AI models — profiled this week by Rest of World — represent what the post-hype market actually rewards: smaller models, lower costs, specific tasks, measurable ROI. Sarvam and Krutrim aren't trying to build AGI. They're trying to make customer service work in twelve languages at a cost that makes sense for a country where per-capita GDP is $2,500. That's not as exciting as GPT-5. It's considerably more likely to generate sustainable revenue.

Business impact

Winners: API-first companies and vertical AI. If the big labs are deprioritizing consumer products, the API layer becomes the real product. Companies building specialized applications on top of foundation models — legal AI, medical coding, financial analysis — benefit from continued model access without competing with the lab's own consumer offering. The GPT-4o situation actually helps API customers: less consumer traffic means more capacity and potentially better pricing.

Winners: India's frugal AI ecosystem. With valuations normalizing, investors are looking for capital-efficient AI plays. Indian AI companies building smaller, task-specific models for massive markets represent exactly that. Expect more Western VC interest in non-US AI companies that prioritize efficiency over scale.

Losers: Consumer AI startups. If OpenAI and Anthropic are both pivoting toward enterprise and government, they'll increasingly compete directly with the startups that built on the assumption the big labs would stay consumer-focused. The TAM for "better ChatGPT wrapper" just shrank.

Losers: AI safety-washing. The Anthropic contradiction — too risky for defense, perfect for banking — exposes the fragility of safety branding as a differentiator. When your safety credentials get selectively cited by different government factions, "responsible AI" becomes a positioning statement rather than a technical guarantee.

The uncomfortable middle: power users. Enthusiasts who built workflows around specific models like GPT-4o are discovering that consumer access to specific models is a privilege, not a right. The Reddit thread about finding GPT-4o through third-party API aggregators is a preview of a future where sophisticated users route around consumer product decisions — paying more for what used to be free.

What this means for you

If you're building on AI APIs: Lock in your model access agreements now. The GPT-4o consumer removal is a reminder that model availability is a business decision, not a technical constant. Negotiate contractual guarantees for model access and deprecation timelines. Test your applications against multiple providers — model portability is no longer optional.

If you're an enterprise buyer: The Anthropic-Washington contradiction means you cannot rely on government signals to de-risk vendor selection. Build your own evaluation framework. And watch for which labs prioritize enterprise support infrastructure (SLAs, compliance certifications, dedicated support) versus which are just selling API keys with a bigger logo on the invoice.

If you're investing: The valuation reset is a buying opportunity if you can identify companies with actual revenue, not projected TAM. Look at India's frugal AI space — Rest of World's reporting on Sarvam and Krutrim is a good starting point. Avoid anything whose pitch deck still says "platform" without specifying which customers are paying.

If you're a developer choosing tools: Stop optimizing for the most powerful model. Start optimizing for the most reliable model access at the cost point your application can sustain. The future looks like a mix of providers, not a single winner — build accordingly.

Sources & further reading

Read more