The former NSA Director joined OpenAI's board 4 months after retirement. The people writing AI safety rules spent their careers building the surveillance infrastructure those rules are supposed to govern.
4 moGap: NSA Director retired → OpenAI board
15+Senior intel/military officials in AI companies
$1–5MUS EO 14110 compliance cost per framework
10²⁵EU AI Act FLOP threshold (calibrated to incumbents)
Choose your depth. The data doesn't change — just the explanation.
The people making the rules for AI used to work for the spy agencies. Now they work for the AI companies. And the rules they're writing make it very expensive to compete with those companies.
The NSA director who oversaw America's most powerful surveillance programs joined OpenAI's board 4 months after retiring. He governs the committee that decides OpenAI's safety policies. This pattern — intelligence officials moving into AI governance — is systematic. The AI regulations being written in Washington impose costs ($1-5M per compliance framework) that are trivial for frontier labs but lethal for open-source alternatives.
The EU AI Act 10^25 FLOP threshold was calibrated in 2023-2024 when GPT-4 was estimated at ~2×10^24 FLOPs. This places the 'systemic risk' designation exactly where open-source frontier models would land as they catch up to closed labs — protecting incumbents' position at the capability frontier. Nakasone's appointment to OpenAI's Safety and Security Committee was confirmed June 2024; the SSC governs all of OpenAI's safety and security practices.
EU AI Act (full text): eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689. US EO 14110: federalregister.gov/d/2023-24283. OpenAI board composition: openai.com/about. Nakasone biography: en.wikipedia.org/wiki/Paul_M._Nakasone. EU AI Office registrations: digital-strategy.ec.europa.eu/en/policies/ai-office.
The Personnel Cards
These are real appointments, real timelines. The revolving door between intelligence apparatus and AI governance is systematic — not incidental.
Gen. Paul Nakasone
Left: NSA Director + US Cyber Command (Feb 2024)
Joined: OpenAI Board of Directors (June 2024)
4 months later
Governs OpenAI's Safety and Security Committee — which oversees all safety and security practices at OpenAI.
Anna Makanju
Left: NSC / State / Pentagon (Obama administration)
Joined: OpenAI VP Global Impact
Immediate transition
Described as having "orchestrated OpenAI's DC lobbying strategy." Bridges federal policy apparatus and corporate AI agenda.
Gen. Keith Alexander
Left: NSA Director + CYBERCOM Commander (2014)
Joined: Founded IronNet Cybersecurity
Immediate
Denied NSA surveillance scope to Congress. Founded a surveillance-adjacent AI security firm capitalizing on government relationships built in uniform.
The Pattern
Left: 15+ senior intelligence/military officials
Joined: Senior AI company positions
Systematic
Not isolated coincidences. A pipeline from intelligence apparatus to AI governance — shaping the rules they spent careers circumventing.
The Regulatory Math That Protects Incumbents
When the people writing rules have deep personal and financial ties to specific companies, the rules tend to create moats for those companies. The EU AI Act and US EO 14110 are case studies.
⚙️ The FLOP Threshold Problem
The EU AI Act's 10²⁵ FLOP "systemic risk" threshold was calibrated in 2023–2024 when GPT-4 was estimated at ~2×10²⁴ FLOPs. Open-source models are currently below this threshold. As they improve, they'll cross it — at which point compliance costs ($1–5M) become lethal for open-source collectives. Frontier labs are already past this threshold and have compliance teams. The rule protects the position of whoever is ahead when the line is drawn.
EU AI Act: FLOP Threshold vs. Model Sizes
10²⁵ threshold (log scale). Compliance costs: trivial for OpenAI, existential for open-source.
Compliance Cost Impact by Organization Type
$1–5M EO 14110 compliance cost vs. annual budget. Source: AI Regulation & Incumbency Analysis, June 2026.