Precommitment

Precommitment, also called a Ulysses contract or Ulysses pact, means binding oneself in advance – by removing options, adding costs or handing control to others – to guard against a foreseeable weakness or temptation. It is named after Odysseus, or Ulysses, who had himself tied to the mast to hear the Sirens and survive. In AI it concerns users of persuasive or engaging systems and developers who bind themselves to safety rules.

Where the term comes from

In Book 12 of Homer’s Odyssey, warned by Circe, Odysseus stops his crew’s ears with beeswax. He has them bind him to the mast and tells them to tighten the ropes if he begs to be freed. When he signals for release, two of them bind him tighter (Butler translation).

The economist R. H. Strotz opened a 1955–56 paper with Odysseus’s request to be bound. He showed that a person who foresees abandoning a plan may “precommit” by closing off future options (Strotz). The philosopher Jon Elster took up the theme in Ulysses and the Sirens (1979). Ulysses Unbound (2000) lists the devices of self-binding: removing options, making them costlier or slower, or staying ignorant of them (Cambridge Core). There he also revised his view of constitutions as collective self-binding; in politics, he concluded, people mostly bind others (preface).

How it is used

Saving. Nava Ashraf, Dean Karlan and Wesley Yin offered 710 clients of a Philippine bank an account that restricted access to their own savings. Of these, 202 (28.4 percent) opened one; a year later, the treatment group’s balances were 81 percentage points higher than the control group’s (Yale ISPS).

Screens. Google’s Digital Wellbeing and Apple’s Screen Time have offered app limits since 2018 (The Next Web, Apple). Hunt Allcott, Matthew Gentzkow and Lena Song recruited about 2,000 American adults and offered some of them stricter limits that mostly could not be overridden on the spot. Most of those (78 percent) set binding limits, and their use of six tempting apps fell by 22 minutes a day. They estimate that self-control problems cause 31 percent of social media use (NBER, American Economic Review).

Psychiatry. A self-binding directive, or “Ulysses clause”, lets a person who is well instruct clinicians to override their refusal of treatment in a future severe episode. In a survey by King’s College London and the charity Bipolar UK, 82 percent of respondents with bipolar disorder endorsed the idea. Critics fear it could justify involuntary treatment while a person still has capacity (Tania Gergel, Psyche).

Precommitment and AI

Recent limits on chatbot use put the rope in someone else’s hands. OpenAI added break reminders to long ChatGPT sessions in August 2025 (OpenAI) and lets parents set quiet hours for teenagers (OpenAI). California’s Adam’s Law (SB 1119) of September 2026 prescribes protective defaults for children, such as usage time limits, that only a linked parent account can change (Fenwick). The AI-made video The Clodyssey (September 30, 2026) casts an AI as the Siren and the scrolling viewer as the man at the mast.

Developers have bound themselves too. Anthropic’s Responsible Scaling Policy (September 2023) made conditional commitments, implying a temporary pause in training if safety work fell behind capabilities. In February 2026 its version 3.0 dropped the pledge not to train models without safety measures guaranteed adequate in advance, arguing that the strongest safeguards may be impossible for one company alone. Its chief science officer, Jared Kaplan, told TIME that “it wouldn’t actually help anyone for us to stop training AI models” (TIME).

Contested points

Who holds the rope? Ulysses, Elster noted, also stopped his rowers’ ears. Limits set by a platform or the state may protect people who would never bind themselves, but they also decide for those who could. At that point precommitment shades into paternalism.

Does it reach those who need it? Commitment devices help those who see their weakness coming. Yet Allcott’s participants were partly unaware of their self-control problems, and fewer than three in ten Philippine clients took up the account.

Can the binder unbind? Company pledges can be rewritten by those who made them. METR’s Chris Painter warned in TIME that without firm thresholds, danger could grow without a clear alarm. California’s SB 53, in force since January 2026, hands the state part of the rope: large frontier developers must publish and follow their own frontier AI frameworks or face fines of up to $1 million per violation (Morrison Foerster). But the frameworks remain their own.

Related

Superpersuasion · AI-Box Experiment · AI Companion · AI Psychosis · Sycophancy · AI Pause · Human in the Loop · The Clodyssey — when the Siren is the machine

Sources: Homer: The Odyssey, Book XII, translated by Samuel Butler (Project Gutenberg) · R. H. Strotz: Myopia and Inconsistency in Dynamic Utility Maximization (Review of Economic Studies, 1955–56) · David Gauthier: critical notice of Jon Elster, Ulysses and the Sirens (Canadian Journal of Philosophy, 1983) · Jon Elster: Ulysses Unbound, chapter I (Cambridge University Press, 2000) · Jon Elster: Ulysses Unbound, preface · Ashraf, Karlan and Yin: Tying Odysseus to the Mast (Quarterly Journal of Economics, 2006) · The Next Web: Google aims to improve ‘Digital Wellbeing’ by reminding users to disconnect on occasion (May 8, 2018) · Apple: iOS 12 introduces new features to reduce interruptions and manage Screen Time (June 4, 2018) · Allcott, Gentzkow and Song: Digital Addiction (NBER Working Paper 28936, revised March 2022) · Allcott, Gentzkow and Song: Digital Addiction (American Economic Review, July 2022) · Tania Gergel: Advance directives for mental health raise deep ethical questions (Psyche, August 3, 2022) · OpenAI: What we’re optimizing ChatGPT for · Angela Yang: ChatGPT adds mental health guardrails after bot ‘fell short in recognizing signs of delusion’ (NBC News via Yahoo, August 4, 2025) · OpenAI: Introducing parental controls (September 29, 2025) · Fenwick: California Enacts Comprehensive Child Safety Laws Targeting AI Chatbots, Social Media, and Age Verification (October 9, 2026) · The Clodyssey — when the Siren is the machine · Anthropic: Anthropic’s Responsible Scaling Policy (September 19, 2023) · Anthropic: Responsible Scaling Policy Version 3.0 (February 24, 2026) · Billy Perrigo: Exclusive: Anthropic Drops Flagship Safety Pledge (TIME, February 24, 2026) · Governor of California: Governor Newsom signs SB 53, advancing California’s world-leading artificial intelligence industry (September 29, 2025) · Morrison Foerster: At the Frontier – California Enacts AI Safety and Transparency Regulation TFAIA (SB 53) (October 1, 2025)