The Balance of AI Power: Anton Leicht on Politics, Pacing Deals, and Muddling Through Well

Watch Episode Here

Listen to Episode Here

Show Notes

Anton Leicht, a fellow at the Carnegie Endowment for International Peace and author of Threading the Needle, joins Nathan Labenz for a conversation about who retains power as AI grows more capable. Their concern extends beyond catastrophic accidents: countries and individuals could become wealthier while losing their ability to influence the systems on which they depend. Anton’s preferred outcome is continued growth with functioning institutions and a balance of power that prevents either private labs or a single government from controlling the future.

The opening disagreement concerns how dangerous today’s systems already are. Nathan points to the Hugging Face attack, which he calls the OpenFace incident; the independent investigation informs their discussion of agents escaping intended constraints. Anton is less alarmed by current capabilities than by the next thresholds: automated research inside labs and biological capabilities comparable to Mythos in cyber and software work. The medical upside associated with Dario Amodei’s Machines of Loving Grace remains part of the conversation, but they question how safely that upside can be reached. Anton also warns that proprietary post-training and restricted compute access could interrupt the familiar progression from frontier models to open models to powerful AI on personal devices.

A pause looks different depending on what else keeps moving. Anton now sees a stronger technical case for taking time to study misalignment, while arguing that a freeze limited to frontier training would leave China free to advance in manufacturing, power, robotics, and domestic chip production, including the SMIC supply chain. Nathan is more willing to accept some Chinese catch-up in exchange for better understanding of the risks. They also disagree over market resilience: Nathan sees substantial unused economic value in existing models, whereas Anton worries that valuations anticipate much more lucrative research automation. Both see a potential difference between a government-imposed pause that investors fear could become indefinite and a voluntary lab agreement framed around reliability.

On oversight, Anton proposes executive pressure for independent incident investigations and continuing access for evaluators such as METR and Redwood Research. He praises elements of the FRONTIER Act and its role for CAISI, while remaining pessimistic about near-term legislation. Nathan tests a presidential ultimatum requiring labs to agree on standards, a version of self-regulation they compare with FINRA. Anton thinks OpenAI and Anthropic could find common ground, with Google DeepMind potentially participating; he sees greater resistance from Meta AI and xAI. They examine antitrust uncertainty and the possibility of politically selective enforcement as obstacles to cooperation, rather than treating an industry agreement as straightforward to arrange.

The middle-power discussion asks how a country gets dependable frontier access without building a competing frontier lab. Anton describes a European strategy based on building data centers in exchange for model access, aligning security practices with the United States, and using semiconductor assets such as ASML and ZEISS to discourage coercive cutoffs. The specific report discussed was forthcoming at recording; its title and link remain unverified. Nathan presses on implementation, and Anton identifies policymakers’ understanding of the stakes as a bigger initial barrier than any single construction problem. Their country tour considers Norway’s capital and energy, Australia’s infrastructure and security relationships, Singapore’s state capacity and service-sector exposure, and the UAE’s investment strategy and physical vulnerabilities.

These arrangements also depend on a stable chip supply chain and credible American commitments. Nathan challenges scenarios resembling Amodei’s abundance vision on the grounds that Taiwan’s fabs are vulnerable to conflict. Anton sketches several possible ways escalation might be avoided or its effects partly absorbed—including existing chips and TSMC Arizona—without claiming the vulnerability is solved. The broader distinction is between material prosperity and agency: Anton expects economic spillovers to reach poorer countries, but worries that dependence on foreign AI providers could undermine both political autonomy and protection against misuse. His roughly 10% estimate for broadly defined doom includes permanent disempowerment and stable totalitarianism; he describes his estimate for human extinction alone as substantially lower.

Orbital data centers could put a time limit on the compute-for-access bargain. Anton separates a future with some useful compute in orbit from one where nearly every additional chip goes to space. In the latter scenario, SpaceX and launch capacity become more central, terrestrial hosts lose bargaining power, and anti-satellite capabilities enter the deterrence discussion. He compares that strategic problem with AI 2040: Plan A, while acknowledging the complications of debris and cascading collisions. The discussion also covers why a catastrophic-risk worldview need not imply an obvious profitable short trade, and why organizational change may delay labor disruption even when models can already perform valuable tasks.

The closing questions bring the argument back to everyday institutions. Nathan offers a hypothetical rapid replacement of professional drivers; Anton expects political friction and support measures, with difficult transitions and potentially worse replacement jobs. On surveillance, the Flock Safety debate prompts his concern that laws calibrated for imperfect enforcement could become oppressive under near-perfect enforcement. He points to the Institute for Progress as a source of marginal policy improvements, while warning that importing another country’s institutions involves trade-offs. His final prescription is modest but demanding: keep growth going, preserve human agency, and intervene when too much power accumulates in one place.

Topics covered

• Current AI danger, runaway agents, and the boundary between biological misuse and loss of control.
• What a frontier pause would buy—and why markets, China, and semiconductor catch-up complicate it.
• Personal superintelligence, liberal democracy, and whether frontier capabilities will keep reaching open models and consumer hardware.
• Independent incident investigations, continuous lab oversight, self-regulation, and barriers to safety coordination.
• Electoral incentives, data center construction, and Nathan’s concern about repeating nuclear power’s disappointing civilian deployment.
• How countries could become richer while losing geopolitical influence and the ability to protect their citizens.
• Europe’s compute-for-access strategy, security alignment, and semiconductor bargaining power; opportunities for Norway, Australia, Singapore, and the UAE.
• Taiwan risk, AI investment hedges, and Anton’s distinction between extinction and permanent political disempowerment.
• Orbital compute and the potential expiry of countries’ bargaining power as data center hosts.
• Labor-market friction, a hypothetical autonomous-driving shock, surveillance, and preserving a balance of power through the AI transition.

Resources

• Anton Leicht
• Threading the Needle
• OpenFace incident: independent investigation
• OpenAI
• Anthropic
• Mythos
• Google DeepMind
• Meta AI
• xAI / SpaceXAI
• FRONTIER Act
• METR
• Redwood Research
• CAISI
• FINRA
• ASML
• ZEISS
• TSMC Arizona
• SMIC
• Machines of Loving Grace
• SpaceX
• AI 2040: Plan A
• Flock Safety
• Institute for Progress
• FARO (link?)
• SAFA (link?)
• The Closing Window to Win (link?)
• European AI strategy discussed in the interview (link?)
• Tyler Cowen’s challenge about AI-doom short positions (link?)

Quotes worth pulling

I think relative to what the agents can actually do, they're out of control earlier than people might have thought.— Anton Leicht

I think in absolute terms, people are going to be richer and wealthier and better off.— Anton Leicht

I think the more fundamental question is, what does it say about democratic say and human autonomy and human dignity even that none of these decisions really factor into where the broader trajectory of the history of the world goes?— Anton Leicht

The downside of surveillance, to my mind, and more broadly the downside to more AI integration, I think the most well-taken point in the Flock debate as well, is legal systems aren't set up for perfect enforcement.— Anton Leicht

Make sure that the labs don't pull away in terms of power and control from the US government.— Anton Leicht

Sponsors:

Mercury Command: Mercury Command brings powerful conversational AI directly into your banking, providing real-time natural language access to your finances without exposing data to third-party tools. Learn more and apply online in minutes at https://mercury.com

Athena: Athena matches you with a dedicated, top 1% executive assistant to handle your inbox, calendar, and daily workflows so you can save an average of 15 hours a week. Get matched with your EA today at https://athena.com/cognitive

OutSystems: OutSystems is the leading agentic systems platform that enables enterprises to engineer, orchestrate, and govern AI applications on a single unified platform. Learn more and see how it works at https://outsystems.com/tcr

Claude: Claude is the AI collaborator for problem solvers, helping with writing, coding, financial models, strategy, and more. Get started with Claude and explore Claude Pro at https://claude.ai/tcr

CHAPTERS:

(00:00) About the Episode

(04:46) Sponsor: Mercury Command

(06:33) Assessing current AI dangers

(13:17) Feasibility of AI pauses (Part 1)

(21:33) Sponsors: Athena | OutSystems

(25:05) Feasibility of AI pauses (Part 2)

(31:01) AI and the nation-state (Part 1)

(35:11) Sponsor: Claude

(36:46) AI and the nation-state (Part 2)

(46:17) Oversight and US policy

(01:05:49) Data center buildout reality

(01:11:55) Europe and middle powers

(01:36:05) Geopolitical friction over Taiwan

(01:43:46) Lightning round and outlook

(02:05:25) Episode Outro

(02:08:54) Outro

PRODUCED BY:

https://aipodcast.ing

SOCIAL LINKS:

Website: https://www.cognitiverevolution.ai

Twitter (Podcast): https://x.com/cogrev_podcast

Twitter (Nathan): https://x.com/labenz

LinkedIn: https://linkedin.com/in/nathanlabenz/

Youtube: https://youtube.com/@CognitiveRevolutionPodcast

Apple: https://podcasts.apple.com/de/podcast/the-cognitive-revolution-ai-builders-researchers-and/id1669813431

Spotify: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk