Global Tech Policy Tracker — 2026-10-01

Top Story

California Bans "Robo Bosses" From Firing Workers — First U.S. State to Codify AI Employment Safeguard

California Governor Gavin Newsom has signed legislation banning artificial intelligence systems from making autonomous decisions to terminate workers, the first state-level law in the U.S. addressing AI in employment decisions. This represents a dramatic reversal from Newsom's previous veto position, reflecting mounting concern over automated workforce management systems that operate without adequate human oversight.

The law requires human review and approval before any employment termination decision, effectively prohibiting algorithms from independently firing workers. This approach distinguishes itself from earlier proposals focused on transparency or testing requirements—instead imposing a hard restriction on autonomous decision-making in a high-stakes employment context.

The significance extends beyond employment law. The policy establishes a new regulatory model: direct prohibition of specific AI applications rather than risk-based frameworks. This contrasts with the EU AI Act's tiered risk approach and federal U.S. discussions centered on transparency and safety audits. California's move signals state-level willingness to restrict AI capabilities outright when employment and economic security are at stake.

The law affects any AI vendor or employer deploying automated termination systems. Compliance requires redesigning systems to route all termination decisions through human decision-makers, likely increasing administrative costs for companies relying on automated workforce optimization.

New Legislation & Regulatory Actions

United States: Sanders-Casar Ban Artificial Superintelligence Act

• What happened: Senators Bernie Sanders and Representative Greg Casar introduced federal legislation requiring federal charter approval for advanced AI model development, with a proposed Cabinet-level AI department empowered to freeze model deployments, order systems offline, and impose penalties of up to 20 years in prison for violations.
• Who it affects: AI developers, frontier model creators (OpenAI, Anthropic, etc.), and large technology companies deploying advanced AI systems
• Status: Introduced October 1, 2026; awaiting committee review
• Why it matters: This represents the most aggressive federal AI governance proposal to date, exceeding previous discussions of safety boards or transparency requirements. It creates criminal liability for AI developers—unprecedented in U.S. AI policy—and centralizes control in a new federal agency, directly conflicting with Trump administration efforts to establish a unified national policy preempting state laws.

Connecticut: AI and Data Privacy Laws Take Effect

• What happened: Multiple Connecticut state laws regulating artificial intelligence, social media use by minors, and facial recognition technology entered into force on October 1, 2026.
• Who it affects: Connecticut-based AI companies, social media platforms targeting minors, and employers using facial recognition
• Status: Effective as of October 1, 2026
• Why it matters: Connecticut joins California and other states in creating enforceable AI guardrails independent of federal action. The laws address workforce AI specifically and restrict biometric data use, signaling state-level enforcement even as federal preemption efforts stall in Congress.

United States: Senate Democrats' AI Safety Bill Blocked

• What happened: Senator Ted Cruz blocked a Democratic proposal to establish a safety board within the Commerce Department to vet AI models before deployment.
• Who it affects: Frontier AI developers, federal oversight framework
• Status: Blocked September 29, 2026; bill did not advance
• Why it matters: Demonstrates partisan gridlock on federal AI governance. The blocked proposal would have created a pre-deployment review mechanism—the most concrete federal safety mechanism attempted. Its failure means the U.S. continues operating without federal model-vetting infrastructure, even as the EU AI Act imposes conformity assessments and certification requirements.

Enforcement & Penalties

• EU Regulators → Pan-European Recruitment Software Firm: A €15 million fine was imposed on a prominent AI-powered candidate-screening system for failing to meet transparency and data governance requirements for high-risk AI systems under EU AI Act Articles 13–15. This represents the first major enforcement action against a specific AI application post-implementation, setting precedent for recruitment AI compliance.

• Italian Data Protection Authority → OpenAI: Italy fined OpenAI €15 million for GDPR violations in training data processing, establishing that regulators expect documented controls, technical safeguards, and evidence of compliance—not aspirational ethics statements alone.

Industry Response

• AI companies and legal counsel: Industry responses focus on dual compliance with fragmented U.S. state laws and the EU AI Act, as federal preemption efforts have stalled. Companies are establishing separate legal and technical compliance programs for California (employment AI restrictions), Connecticut (facial recognition bans), and EU markets (high-risk conformity assessments). Legal guidance emphasizes that the Sanders-Casar proposal, if enacted, would require fundamental restructuring of model deployment timelines and governance.

• Trump administration and corporate AI leaders: On September 29, Trump announced an accord signed by major AI companies committing to self-policing measures, including robust safeguards against rogue models, external audits, and independent board committees. However, critics note the accord lacks enforcement mechanisms and contrasts sharply with the Sanders-Casar bill's mandatory federal oversight.

• Technology policy analysts: Industry observers warn that the fragmented regulatory landscape—state bans on specific AI applications, EU conformity assessments, and proposed federal criminal liability—creates compliance costs that favor large enterprises while constraining startup innovation. The €35 million-or-7%-of-revenue penalty ceiling under the EU AI Act now faces pressure from state-level restrictions with per-incident fines (Connecticut: up to $100,000 per violation).

Region Scorecard

Analysis: What This Means

• Dual compliance becomes mandatory for global AI vendors: Companies building AI systems for hiring, recruitment screening, or autonomous decision-making now must comply with California's hard ban on autonomous termination, Connecticut's restrictions on data use, the EU's high-risk AI conformity framework (with €35M penalties), and potential federal criminal liability if the Sanders-Casar bill advances. This fragmentation makes it economically rational for large enterprises to build separate compliant systems per jurisdiction while smaller vendors may exit some markets entirely.

• Enforcement is shifting from guidance to penalties: The €15M fines against recruitment AI and OpenAI signal that regulatory agencies are moving from compliance encouragement to active enforcement. Regulators now expect documented technical controls (transparency logs, external audits, data governance records) and will impose penalties for documented gaps. Companies should prioritize third-party audits and compliance documentation over aspirational policy commitments.

• Federal AI policy gridlock accelerates state-level action: With Cruz's block of the Commerce Department safety board and the Sanders-Casar bill's aggressive posture unlikely to pass, states are filling the void. Expect California, Connecticut, and other large states to introduce sector-specific AI restrictions (employment, housing, insurance, credit) throughout 2026, creating de facto national standards through market size pressures.

• The "self-policing accord" lacks credibility with regulators: Trump's announced industry accord relies on voluntary disclosure and internal audits, but recent EU enforcement actions and proposed federal criminal penalties signal regulators expect independent external oversight, not corporate self-reporting. Companies should treat the accord as insufficient and prepare for external regulatory audits regardless.

What to Watch Next Week

• October 2–5: Expected EU announcement on high-risk AI system conformity assessment timelines and first-wave enforcement targets. Watch for additional fines against facial recognition, credit-scoring, or hiring systems.

• October 3: Congressional hearing on Sanders-Casar bill preliminary feedback. Industry and civil liberties testimony will shape whether the proposal gains traction or faces committee delays.

• October 7: Deadline tracking for any announced state-level follow-ups to California and Connecticut. Several states (Illinois, Massachusetts) may announce AI employment or bias-audit laws entering 2027.