The Download: controlling AI agents and “Super Intelligence” loyalty tests

Glass Epoch DeskOriginal digestSource: MIT Technology Review

AI-assisted digest, checked against the original source before publishing. How we work

Today's Download covers "reverse game theory" as a lever for agent control, plus the Trump administration's "Super Intelligence" rebrand and a roundup of tech stories.
The Download: controlling AI agents and “Super Intelligence” loyalty tests
Photo: Frantzou Fleurine · free stock license

Researchers behind the idea offer a splitting-a-cake analogy: if one person cuts the cake and the other chooses which slice to take first, the cutter is motivated to divide it evenly. The hope is that similarly constructed incentive structures could push large models to follow human instructions and constraints more reliably. The newsletter is explicit that open questions remain about whether such mechanisms hold up in practice, and it links to a fuller piece on how the approach could work.

The second lead item is political. At the end of September, President Trump signed an executive order carrying a single directive: federal agencies must rebrand artificial intelligence as "Super Intelligence." One month in, the new terminology is showing up gradually across government sites and documents, though it has not spread everywhere — several major tech companies still say "AI." The stated concern is bigger than vocabulary: use of the phrase is emerging as a loyalty signal to the administration, and declining to use it can invite attacks.

The rest of the edition is a ten-item must-read roundup drawn from other outlets. Axios reports that candidates in today's US midterms are betting AI regulation will move voters, with investors watching the results closely. The Financial Times reports Nvidia has told data-center clients that the majority of their sales must be non-Chinese, a move following pressure from Washington. CNBC and BBC cover European attacks that turned supermarket refrigerators into a food-security concern and raised alarms about other appliances.

Other entries range widely: signals detected at the smallest scale yet explored via CERN's Large Hadron Collider; brain stimulation that can sharpen performance while reducing motivation; big tech spending on gas plants to power data centers; Google saying an Android partner infected 1.3 million devices; Apple's difficult first-robot effort; AI tokens as a Singles' Day product category in China; and a Sahara meteorite offering clues to Mars. The quote of the day comes from family lawyer Steven Benmor, who tells the Wall Street Journal that clients increasingly face debates between human and robot experts.

What to watch: whether reverse-game-theory incentives survive contact with real model deployments, or remain a theoretical alignment sketch; whether the "Super Intelligence" mandate hardens into a genuine government-wide terminology shift or stalls with major firms; and how today's midterms, plus Nvidia's China restrictions, reshape the regulatory and market environment builders operate in.

Key facts

  • • Trump signed an executive order late September ordering an AI "Super Intelligence" rebrand.
  • • Nvidia told data-center clients the majority of sales must be non-Chinese.
  • • Google said an Android partner infected 1.3 million devices with malware.
  • • CERN's Large Hadron Collider detected signals at the smallest scale yet explored.
  • • US midterm candidates are betting AI regulation will sway voters.

Why it matters

Builders shipping agents face a new control paradigm — designing incentive rules rather than hard constraints — but its practical reliability is unproven. On the policy side, terminology itself is becoming a political marker in US federal AI work, while Nvidia's China sales limits and midterm outcomes shape near-term market and regulatory conditions.

Spotted an error? Send a correction.

Get the day's AI news in one email

The Daily Epoch · 60-second read · unsubscribe anytime

Related

Policy
WIRED AI3 min read

Should AI Decide Your Court Case? This Judge Says Yes

Outgoing Master of the Rolls Geoffrey Vos argues AI could adjudicate routine civil cases — while the profession wrestles with hallucinated citations, sanctions, and a flood of chatbot-assisted claims.