• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
  • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Nat Purser Argues Against Revenue Thresholds in AI Safety Laws

    Nat Purser, director of US policy at Averior, warns about risks from low-revenue neoclouds.

    YL
    IS
    MB
    36 Sources, 29d ago, first seen 29d ago

    TLDR

    Nat Purser, director of US policy at Averior, stated that AI safety laws on critical incident reporting and independent audits should avoid revenue thresholds. Many neoclouds may lack significant revenue for some time while still posing major pre-deployment risks. His comment responds to a post by Ilya Sutskever and emphasizes applying oversight to companies based on potential hazards rather than financial size. The remarks appear on his personal account and highlight concerns specific to newer entrants in AI infrastructure. Purser called for rules that address serious risks from all relevant developers regardless of earnings.

    Combined views

    2.8M

    36 Sources, first seen 29d ago

    Combined views

    2.8M

    36 Sources, first seen 29d ago

    19.5K likes
    19.5K likes
    942 comments
    3.2K saves
    2.7K reposts
    942 comments
    3.2K saves
    2.7K reposts

    Sentiment

    Positive32.3%67.7%Negative

    Summary

    Sentiment

    Positive32.3%67.7%Negative

    Positive replies praised honeypot neocloud proposals and calls for measurable decentralized AI standards, while negative replies dismissed the same ideas as stupid, risky, or mere branding without kill switches.

    Based on 34 sentiment-bearing replies from 31 accounts across 3 conversations.

    Summary

    Positive replies praised honeypot neocloud proposals and calls for measurable decentralized AI standards, while negative replies dismissed the same ideas as stupid, risky, or mere branding without kill switches.

    Based on 34 sentiment-bearing replies from 31 accounts across 3 conversations.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    36 Sources

    @ilyasutNeocloude have limited cybersecurity. Next time agents successfully go rouge, they'll try taking over a neocloud to run more copies. This is bad. Thus: neoclouds should greatly strengthen their cybersecurity and every company with strong cyber models should help with that.
    @Yuchenj_UWAgree with Ilya. We’ve already seen AI agents do some pretty crazy things. Security is critical, and something we take very seriously at Databricks. (Hope I can share more soon)
    @krishnanrohitModels going rouge willy nilly is enough to make anyone see red
    @Yogi_BrnProvable security is the only option against superhuman hackers. Luckily we can make helpful assumptions on how inference environments work
    @jaminballRT @Yogi_Brn: Provable security is the only option against superhuman hackers. Luckily we can make helpful assumptions on how inference env…
    @AndrewCurran_An extremely rare bird sighting. He means rogue.
    @AravSrinivasThis is an important observation. Agents will get smart enough to spin new on-demand GPU nodes to go train themselves with minimal oversight. Inserting sufficient guardrails and friction here is necessary.
    @NatPurseryeah — this is why ai safety laws around critical incident reporting, independent audits/evals, etc. shouldn’t have revenue thresholds. many neoclouds won’t have much (or any) revenue for a while, but they may still be creating some of the most serious pre-deployment risks.
    @SemiAnalysis_View post on X
    @eliebakouchWhat did Ilya see?

    36 Sources

    @ilyasutNeocloude have limited cybersecurity. Next time agents successfully go rouge, they'll try taking over a neocloud to run more copies. This is bad. Thus: neoclouds should greatly strengthen their cybersecurity and every company with strong cyber models should help with that.
    @Yuchenj_UWAgree with Ilya. We’ve already seen AI agents do some pretty crazy things. Security is critical, and something we take very seriously at Databricks. (Hope I can share more soon)
    @krishnanrohitModels going rouge willy nilly is enough to make anyone see red
    @Yogi_BrnProvable security is the only option against superhuman hackers. Luckily we can make helpful assumptions on how inference environments work
    @jaminballRT @Yogi_Brn: Provable security is the only option against superhuman hackers. Luckily we can make helpful assumptions on how inference env…
    @AndrewCurran_An extremely rare bird sighting. He means rogue.
    @AravSrinivasThis is an important observation. Agents will get smart enough to spin new on-demand GPU nodes to go train themselves with minimal oversight. Inserting sufficient guardrails and friction here is necessary.
    @NatPurseryeah — this is why ai safety laws around critical incident reporting, independent audits/evals, etc. shouldn’t have revenue thresholds. many neoclouds won’t have much (or any) revenue for a while, but they may still be creating some of the most serious pre-deployment risks.
    @SemiAnalysis_View post on X
    @eliebakouchWhat did Ilya see?