Cryptographer Matthew Green: AI Bug Hunting Nears a Ceiling; Crypto Backdoors' Biggest Losers Are US Citizens

On August 14, cryptographer Matthew Green posted a series of messages on X outlining the security and policy consequences of AI's leap in vulnerability discovery: the low-hanging fruit of AI bug-hunting is nearly exhausted and a 'vulnerability ceiling' exists; governments may respond by explicitly demanding access backdoors to encrypted systems and devices, with American citizens as the biggest losers; and open-weight models will inevitably be released by other nations, shortening every exploit's lifespan and ending the current fragile 'ceasefire.'

Confirmed

• Green acknowledged that AI systems such as Anthropic's Mythos, OpenAI's models and GLM are impressive at finding vulnerabilities, but the attacker's edge is unlikely to last: low-hanging fruit will run out, and frontier models will only stay ahead of open models temporarily.
• He predicts governments will explicitly demand access to encrypted systems and personal devices like phones; even secretly added backdoors are unlikely to remain undetected, and the world is accelerating 'into darkness.'
• On governments demanding 'explicit access mechanisms,' Green judges the biggest losers to be American citizens rather than foreign targets: other countries would simply ban such backdoors or refuse US products, so the backdoors would only be effective inside the US and its allies.
• Frontier labs might temporarily sustain the fragile 'ceasefire' by giving intelligence agencies preview access to their models, but cannot stop other countries from releasing open-weight models; each vulnerability's usable lifespan will then shrink and attack costs rise.
• He also commented in a repost that AI may soon find vast numbers of software vulnerabilities and make software genuinely secure for the first time — but once software is secure, the current fragile ceasefire will have to end.

Why it matters

• Green connects AI capability, backdoor politics and open-model diffusion into one coherent scenario: if AI massively accelerates vulnerability discovery, legislative demands for 'guaranteed access' and the spread of open weights will simultaneously upend the existing security order.
• The claim that backdoors' biggest victims are domestic citizens adds a geopolitical lens to the crypto backdoor debate: backdoors have limited effect on foreign targets, while the cost is borne mainly by users in the US and its allies.

2026-08-14 ~ 2026-08-14 · 8 related posts

• AI Bug Hunting May Hit a Ceiling: Expert Says Low-Hanging Fruit Will Run Out — matthew_d_green · 2026-08-14
• AI Could Make Software Secure for the First Time, but Fragile Ceasefire May End — matthew_d_green · 2026-08-14
• Open models will shorten software exploit lifespan: Cryptographer — matthew_d_green · 2026-08-14
• [source] AI Security Expert Predicts Government Demands for Backdoors in Encrypted Systems — matthew_d_green · 2026-08-14
• Security Expert: Governments Will Demand Access to Encrypted Systems, 'Going Dark' Accelerates — matthew_d_green · 2026-08-14
• [source] Cryptographer Matthew Green: encryption backdoors will hurt US citizens most — matthew_d_green · 2026-08-14
• Security expert warns AI backdoors will hurt domestic citizens most — matthew_d_green · 2026-08-14

1 near-duplicate retellings: matthew_d_green