The era of unregulated AI is over, and most builders haven’t updated their mental model yet.
From AI layoffs and frontier laws to the boring questions founders now have to answer”
A lot of builders still talk about AI as if the rules have not caught up yet.
That mental model is already out of date.
On the labour side, the numbers have moved from hypothetical to measurable. In 2025 there were about 1.2 million layoffs in the US; roughly 55,000 of them – around 4–5% – were explicitly attributed to AI. In Q1 2026 alone, tech firms cut ~80,000 jobs, with about 47.9% of those layoffs reported as “replaced by AI.” By mid‑2026, one tracking firm estimates that AI‑driven cuts account for ~16% of all recorded layoff announcements.
So AI is no longer just a future shock in slide decks. It is already in the HR spreadsheets, severance budgets and board minutes.
At the same time, the legal environment is hardening at the top end.
California’s SB 53 – the Transparency in Frontier Artificial Intelligence Act – applies to developers training frontier models with more than 10²⁶ FLOPs and earning more than $500 million a year. Those developers must publish safety disclosures, maintain governance structures, and run an ongoing process for spotting and mitigating catastrophic risks.
New York’s RAISE Act uses similar thresholds –
frontier models above 10²⁶ FLOPs and developers over $500 million revenue – and adds requirements for safety protocols, transparency, and fast incident reporting (on the order of 72 hours), backed by civil penalties that can reach millions of dollars for repeat violations.
Across the Atlantic, the EU AI Act is moving from concept to calendar. It entered into force on 1 August 2024 and becomes fully applicable on 2 August 2026, with prohibited practices enforceable earlier and high‑risk system obligations phased through 2026–27 and beyond. Transparency duties for some systems and content labeling obligations start to bite within the next year, even after some deadlines were recently pushed back.
There is still no single “AI law”.
Instead there is something more awkward: a growing pile of overlapping duties that most small teams have not really built into how they work.
The change is not only legal. It is operational.
If you are building something that touches hiring, safety, physical infrastructure, or other high‑risk decisions, the question is no longer just “does the model work?” It is also:
Who decides an AI incident has happened?
What gets logged and where?
What would you send a regulator, customer, or insurer within 72 hours if something goes wrong?
How does that process show up in procurement, security reviews, and your own internal governance?
This is not a policy seminar problem.
It is product, process and trust.
Many founders still think these laws only matter for five to eight frontier labs at the very top of the stack. But even analyses of SB 53 point out that companies below the compute or revenue thresholds will feel indirect pressure as investors, large customers and partners start treating those standards as the “normal” way serious teams operate. New York’s RAISE Act and the EU AI Act will have the same cascading effect through supply chains, enterprise contracts and audits.
In that sense, the unregulated era is not ending with one big bang.
It is ending through procurement questionnaires, incident clauses, and the slow realization that AI systems are being treated more like other safety‑relevant technologies.
The winners in this next phase will not just be the teams with better models or better demos.
They will be the teams that can answer boring but critical questions cleanly:
failure modes,
incident playbooks,
sensor and data quality,
override paths, and
evidence that the system can be monitored and shut down safely.
Most builders have not updated their mental model yet.
They should.
For a deeper dive on how this connects to safety incidents, sensors, infrastructure and the macro story around AI and GDP, there is a longer piece here:
AI, Safety and GDP: The Blind Spot in Today’s Macro Story – over on Intelligent Founder.