Europe's AI Act Is Now Law: What Changes for Developers and Startups
The world's most comprehensive AI regulation takes full effect, creating new obligations for model providers and a potential template for global governance.
On February 2, 2026, the EU's AI Act entered full force, ending a two-year phase-in and beginning the world's most ambitious AI regulation experiment. The law classifies AI systems by risk level and imposes corresponding obligations on every company deploying AI in the EU's 450 million-person market.
The Risk-Based Framework
The Act divides AI into four tiers. Unacceptable risk systems are banned: social scoring, real-time public biometric ID, manipulative AI. High risk systems — healthcare, employment, law enforcement — face extensive requirements: risk assessments, documentation, human oversight, accuracy monitoring. Limited risk systems like chatbots must disclose they're AI. Minimal risk applications face no specific obligations.
General-Purpose AI Rules
The most debated provisions govern general-purpose AI models like GPT-4, Claude, and Gemini. Providers must maintain technical documentation, implement copyright compliance, and provide training data summaries. Models posing "systemic risk" — trained above 10^25 FLOPs — face adversarial testing and incident reporting requirements.
"The GPAI provisions are where the rubber meets the road," says policy advisor Dr. Kai Zenner. "If you offer an API developers build on, you have real compliance work to do."
Impact on Startups
Reactions are mixed. A European Digital SME Alliance survey found 62% of AI startups expect costs to rise at least 15%. Some welcome the clarity; others worry about burden on smaller companies. Regulatory sandboxes and reduced SME fees help, but may not be enough.
"We've hired two full-time compliance officers," says Ada Health co-founder Léa Steinacker. "For a 50-person startup, that's significant."
The Brussels Effect
Despite competitiveness concerns, the Act is influencing regulation worldwide. Brazil, Canada, Japan, and India have introduced legislation drawing on the EU's framework. Columbia Law's Dr. Anu Bradford calls it the "Brussels Effect" — EU rules becoming the de facto global standard.
The US remains an outlier with no comprehensive federal AI legislation, leaving a patchwork of state laws that many find more confusing.
Enforcement Questions
The European AI Office has fewer than 150 staff to oversee thousands of AI systems. Penalties are steep — up to €35M or 7% of global revenue — but whether they'll be levied remains uncertain. "The AI Act is only as good as its enforcement," says Zenner. "If it becomes the GDPR of AI — lots of rules, inconsistent enforcement — it will fail."
More from Pulse AI
Microsoft's Azure Trust Crisis: A Former Insider Exposes the Cracks Beneath the Cloud
A former Azure Core engineer's Substack series alleges years of internal dysfunction at Microsoft nearly cost the company its relationship with OpenAI and eroded U.S. government trust — a damaging account that lands just as Microsoft launches three new in-house AI models designed to compete with its own partners.