Infrastructure leverage: where lawful control actually sits How training clusters, model weights, deployment services, inference systems, and essential-service fallbacks create different control points.
Source-checked 6 min
05 Response
Response options, in order of leverage
This chapter covers lawful civic, institutional and infrastructure measures that can reduce risk. Their effectiveness and tradeoffs vary; none is a universal control point.
Infrastructure-layer leverage
- Compute and deployment controls. Frontier training depends on concentrated chip, cloud and energy supply chains, while copied weights and distributed inference are harder to control. Policy must distinguish those stages.
- Reversible deployment. Organizations should test privilege revocation, credential rotation, network isolation, rollback, model withdrawal and human escalation instead of relying on one vaguely defined “kill switch.”
- Degraded operating modes. Where safety and engineering permit, essential power, water, communications, and payment services should test how they continue when an AI component or provider is unavailable.
Institutional and legal leverage
Where pressure works
Organized, evidence-based pressure on concentrated model, cloud, and compute providers can have more leverage than an isolated consumer choice. Shareholder advocacy, procurement requirements from major customers, coordinated regulatory reporting, and legal liability exposure can shape organizational behavior.
- Support whistleblower protections for employees and contractors inside AI labs. Insiders may see serious problems early and need confidential external channels and protection from retaliation.
- Back evidence-based competition and concentration scrutiny. More independent providers can reduce dependency and private power, while faster competition can also worsen pressure to deploy before risks are understood.
Information-layer defense
AI can make persuasive content cheaper and more personalized, although population-scale effects remain uncertain. Practical countermeasures: default skepticism toward emotionally optimized content you cannot source; supporting provenance standards — such as C2PA Content Credentials and cryptographic signing — that can document parts of a file's origin and editing history; and treating any claimed majority-opinion signal as unverified until confirmed independently. Provenance can be stripped and does not prove that a claim is true.
Infrastructure Institutional Informational
In this chapter
Legal and regulatory leverage How liability, competition policy, procurement, audits, incident reporting, and enforceable duties can shape AI development and use, as under the EU AI Act.
Source-checked 6 min
Defending the information layer Collective defenses against AI-enabled fraud, impersonation, manipulation, and information overload without empowering censorship, using C2PA provenance tools.
Source-checked 5 min
Shareholder and corporate governance advocacy Practical ways boards, investors, workers, and customers can improve AI risk governance—and the limits of each lever, from SEC proposals to board committees.
Source-checked 6 min
Grassroots and civic organizing A lawful, ideologically plural guide to public participation on AI policy, from local procurement to legislation and standards, as in California's SB 53.
Source-checked 6 min
Export controls and sanctions as a safety instrument A September 2026 guide to advanced-chip controls, the rescinded AI Diffusion Rule, licensing, evasion, and strategic tradeoffs, plus 2026's Nvidia shift.
Source-checked 5 min
Media literacy and content provenance How Content Credentials, watermarking, detection, source verification, and critical reading fit together—and where each fails, much like a nutrition label.
Source-checked 6 min
Institutional AI-preparedness playbooks What NIST, federal agencies, and cities like Seattle have actually published to help schools, hospitals, governments, and small businesses prepare for AI.
Source-checked 7 min