SI safety in the European Union

Key dates

Introduced

n/a

Signed or adopted

Jun 13, 2024

Effective

Aug 2, 2025

Next step · in 267 days

August 2, 2027

Deadline for general purpose SI models placed on the market before August 2, 2025 to comply

Scope

Providers placing general purpose SI models on the EU market, with added duties for models with systemic risk, presumed above 10^25 FLOP of training compute.

Penalties

From August 2, 2026 the Commission can fine providers of general purpose SI models up to 3 percent of worldwide annual turnover or 15 million euros, whichever is higher.

What it requires

• Notify the Commission within two weeks of reaching or foreseeing the 10^25 FLOP systemic risk threshold
• Keep technical documentation and give information to downstream providers
• Publish a summary of training content and maintain a copyright policy
• Systemic risk models: evaluate the model, including adversarial testing, and assess and mitigate systemic risks
• Systemic risk models: track and report serious incidents to the AI Office
• Systemic risk models: ensure adequate cybersecurity for the model and its infrastructure

Sources

1. Guidelines for providers of general-purpose AI models European Commission · Jul 18, 2025
2. Overview of Guidelines for GPAI Models EU Artificial Intelligence Act (Future of Life Institute)
3. EU AI Act: The AI Office sends its first requests for information to leading AI providers Reed Smith · Oct 26, 2026

Key dates

Introduced

n/a

Signed or adopted

Jul 10, 2025

Effective

Aug 2, 2025

Next step

None scheduled

Scope

Voluntary for providers of general purpose SI models with systemic risk; the Commission and AI Board treat it as an adequate tool for showing AI Act compliance, and signatories include Amazon, Anthropic, Google, IBM, Microsoft, Mistral AI and OpenAI, with xAI signing only this chapter and Meta not signing.

Penalties

No penalties: this measure has no binding enforcement.

What it requires

• Adopt a Safety and Security Framework and notify it to the AI Office
• Identify and analyze systemic risks, including CBRN, loss of control, cyber offense and harmful manipulation
• Set measurable risk acceptance criteria and proceed only when all risks are acceptable
• Use independent external evaluators, with limited exemptions
• Track and report serious incidents and file a Safety and Security Model Report
• Secure the model and maintain whistleblower protections and a healthy risk culture

Sources

1. The General-Purpose AI Code of Practice European Commission · Oct 7, 2026
2. Signing the General-Purpose AI Code of Practice: FAQ European Commission
3. General-Purpose AI Code of Practice Wikipedia

Key dates

Introduced

n/a

Signed or adopted

Jul 8, 2026

Effective

Jul 27, 2026

Next step · in 24 days

December 2, 2026

Generative systems already on the market must meet the Article 50(2) marking duty, and added Article 5 provisions apply

Scope

Amends the AI Act and gives the AI Office direct supervision of SI systems built on a general purpose model by the same provider and of SI systems in very large online platforms and search engines.

Penalties

Allows periodic penalty payments capped at 5 percent of average daily worldwide turnover per day, alongside existing AI Act fines.

What it requires

• Delays stand-alone Annex III high-risk obligations to December 2, 2027
• Delays high-risk obligations for Annex I product components to August 2, 2028
• Gives the AI Office information, inspection, binding commitment and fining powers (Articles 75a to 75c)
• Gives generative systems placed on the market before August 2, 2026 until December 2, 2026 to meet output marking duties
• Does not delay the general purpose model obligations or their August 2, 2026 enforcement start

Sources

1. Digital Omnibus on AI EU Artificial Intelligence Act (Future of Life Institute)
2. EU AI Omnibus enters into force, amending the AI Act White & Case