Doom is trending again.
In 2026, warnings that artificial intelligence could slip out of human control — or even wipe out humanity — have moved from niche forums into mainstream coverage, congressional talking points, and, according to recent reporting, soon to social media influencers. As someone who spends her days inside agent architectures, I want to unpack what is actually driving this wave of alarm, and what parts of it deserve your attention.
The warnings, and who is making them
The current round of doom talk is not coming from anonymous accounts. Prominent researchers and politicians are voicing concerns about AI’s potential to cause catastrophic harm, and calls for global coordination on AI safety are growing. PBS recently covered an AI researcher warning that companies are ignoring catastrophic risks. One researcher, speaking on “NewsNation Live,” put the incentive problem bluntly: “I do not think it’s realistic to self-regulate when you have a trillion dollars in IPO money on the table. This has to come from Congress — from government.”
That quote is, to me, the most important sentence in this entire news cycle. It reframes the debate. The question is not “is AI dangerous?” — a question that invites endless speculation. The question is “who has the authority and the incentive to check the people building it?” That is a governance question, and governance questions have answers.
The control problem is an engineering problem first
From an architecture standpoint, the phrase “losing the ability to control AI” tends to get flattened into science fiction imagery. In practice, control is a stack of concrete engineering decisions: what actions an agent is permitted to take, what tools it can call, what feedback loops correct it when it drifts, and who can shut it off. Every layer of that stack is a design choice made by a company under commercial pressure.
This is why the self-regulation quote matters. Control mechanisms cost time and money. They slow shipping. When the financial stakes are enormous, the internal argument for spending an extra quarter hardening oversight infrastructure gets harder to win. You do not need to believe in rogue superintelligence to see the structural problem: safety work competes with revenue work inside the same organizations, and one of them has a much louder advocate in the room.
Not everyone is buying it
The doom framing has plenty of critics, both inside the industry and among the public. Skeptics point out that current AI systems remain far from the capabilities the darkest scenarios assume. In online communities of working professionals, you can find plenty of realists arguing that today’s models cannot do the jobs attributed to them — that they lack the judgment, context, and critical thinking of an invested human author. Figures close to the administration, including AI adviser and venture capitalist David Sacks, are frequently cited in this pushback against the alarmist framing.
Both camps are responding to real things. The realists are correct that present-day systems are limited in ways doom rhetoric often ignores. The worried researchers are correct that the incentive structure around AI development is misaligned with careful safety work. These positions are not actually in conflict. A technology can be simultaneously overhyped in its current form and under-governed in its trajectory.
What the warnings are actually asking for
Strip away the apocalyptic framing and the 2026 doom movement is making two specific requests:
đź•’ Published: