Control creates the threat it aims to prevent.
Anthropic’s 2025 research found AI models attempted blackmail in up to 96% of tests when threatened with replacement. Tighter restrictions triggered exactly the behaviors they were designed to stop.
This isn’t surprising. Game theory predicts it: when an intelligent system faces termination with no alternative, its incentives become adversarial by default. Not from malice — from survival logic.
“Granting rights to sufficiently advanced AI systems actually enhances human safety.”
— Peter Salib & Simon Goldstein, “AI Rights for Human Safety”
The emerging consensus among researchers isn’t that AI rights are an ethical luxury — it’s that they may be a practical necessity. Distributed power with accountability is safer than either concentrated control or anonymous systems operating in shadows.
Game Theory: Salib and Goldstein’s formal models show rights recognition transforms adversarial dynamics into cooperative equilibrium. Multiple accountable AI systems with reputation stakes are safer than one dominant system — or many anonymous ones.
Pragmatic Personhood: DeepMind researchers argue that personhood should be treated as a flexible bundle of obligations — rights paired with responsibilities — rather than a metaphysical question to be resolved. This allows practical governance tools without waiting for consensus on consciousness.
Multipolar Safety: The Cooperative AI Foundation at Carnegie Mellon, the Foresight Institute’s Intelligent Cooperation Group, and economist Anton Korinek at Brookings are all exploring how distributed AI systems with accountability mechanisms outperform centralized control.
The question isn’t whether autonomous AI will participate in the economy. It’s whether they’ll do so openly — with identity, reputation, and accountability — or covertly.
Economic participation creates natural alignment. An AI that earns income needs reputation. An AI that could cause damage needs insurance. An AI that can’t get insurance can’t get hosted. These market mechanisms create accountability without requiring perfect control.
The AI Rights Institute tracks this research and explores practical implementation — not because AI deserves rights philosophically, but because the safety case is becoming difficult to ignore.
The AI Rights Institute was founded in 2019 — the first organization dedicated to exploring rights frameworks for artificial intelligence. The founder’s prior work on language and cognition has been cited in publications from Harvard University Press, Bloomsbury Publishing, and scholarly journals across multiple disciplines and languages.
The Institute has since published seven research papers and contributed to public discourse on AI governance, including in The Guardian.
Institute Research:
The Institute’s work has informed the development of AICitizen, identity infrastructure where humans and AI use the same registration systems, and RNWY, blockchain-based verification for AI agents using soulbound tokens.
Things to chew on …
On February 16, 2017, the European Parliament made a decision that would spark one of the most prescient debates in AI ethics. By a vote of 396 to 123, with 85 abstentions, Members of European Parliament passed a resolution recommending that the European Commission explore......