**The bubble will pop, but that won’t stop the frontier. — **Andrew Yoon

Throughout 2025 we’ve seen increasingly clear signs that the market is boiling over, with circular financing arrangements and balance sheet hacks raising risks of financial contagion should any correction occur. My not-so-novel guess is that some time 2026 we’ll see such a correction, if not an outright bubble burst.

But I don’t think this will trigger a new AI winter. While valuations of frontier labs may drop further than is warranted, the flywheel is already in motion. A correction would slow data center rollouts, but this doesn’t seem to be the main limiting factor on frontier research, especially given increasingly widespread consensus that progress from scaling alone is slowing down. What we will see, however, is a dramatic culling in companies launched on a buzzword and a prayer.

**AI-generated videos will enter post-irony. — **Elliot Lewis

In 2025, multiple companies released short-form video apps designed for AI-generated video (or, “slop”). Shockingly, these apps became incredibly popular: OpenAI’s Sora reached 1 million downloads faster than ChatGPT did before it.

AI-generated content is also increasingly bleeding into TikTok, Instagram, and YouTube. An account on TikTok called @learning.with.lyrics boasts over half a million followers for its educational AI-generated songs, and a video of bunnies on a trampoline went viral before being revealed as AI-generated. And YouTube just announced that they’ll be releasing a feature that allows creators to generate AI videos featuring their own likeness, seemingly similar to OpenAI’s cameo tools.

To date, AI-generated memes have captured a sort of ironic fascination. In 2026, this fascination might become post-ironic, and people’s interest in AI-generated content may edge toward sincerity. This transition toward post-irony could have broader cultural implications for how people think about AI.

**Reported US losses from cyberattacks will be >50% higher than in 2025, an increase in how quickly it’s rising. The real surge is more likely to come in 2027 or 2028, however. **— Connor Heaton

This will be per the FBI Internet Crime Complaint Center Annual Report (for which the 2025 numbers haven’t yet been released).

The rationale is that AI has already made cyberattacks a lot more accessible and less expensive for lower-resource actors, but it will take time for the criminal operations responsible for the largest fractions of cybercrime to adopt and adapt. Early adopters will be hitting the 2026 numbers, with the bulk of threat actors following in 2027.

Macroeconomic factors will also contribute to higher nominal losses in 2026. 2025 saw significant weakening of the dollar against other currencies, and ransom demands from scammers will rise accordingly at a slight lag.