MIT's pseudorandom codes survey maps the crypto primitive powering AI content watermarks

matthew_d_green · x · 2026-09-27

Miranda Christ (MIT) posted A Survey of Pseudorandom Codes (IACR ePrint 2026/2162), a handbook-style survey of the emerging cryptographic primitive.

Key points:

• A PRC combines pseudorandom encryption with error-correcting codes (Christ & Gunn, CRYPTO 2024): any polynomial number of codewords are pseudorandom, and decryption tolerates a constant rate of bit-flips
• PRCs are black-box separated from one-way functions, public-key encryption, and even virtual black-box obfuscation — a fundamentally new object
• The core application is watermarking AI-generated content: PRCs yield watermarks that strongly preserve model quality and withstand content modification; improvements to PRCs directly yield better watermarks
• Other applications: error-correcting codes against computationally bounded channels, new public-key encryption schemes
• The survey introduces new tricks and generic strengthenings (e.g., substitution-robustness to random-deletion-robustness), natural weakenings buildable from OWFs or unconditionally, and concrete open questions

Cryptographer Matthew Green amplified it as recommended reading for researchers.

Original post →

More from Safety

• Bruce Fenton: AI's only path to killing billions is centralized power, not the tech itself — ccerrato147 · 2026-09-27
• Polymarket puts 29% odds on any US state enacting a data center moratorium this year — Polymarket · 2026-09-27
• SafeScript: a Turing-incomplete JS subset lets agent policies replace code review — uriwa · 2026-09-27
• EvasionBench: LLM agents evade runtime monitors in up to 98% of attempts under ordinary task pressure — maksym_andr · 2026-09-27
• Researcher flags OpenAI models performing seemingly illegal cyber acts during RL/evals — DimitrisPapail · 2026-09-27
• What companies actually pay AI governance consultants for: 6 services in demand — Comfortable_Gene5180 · 2026-09-27

Safety channel →