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

    Academic Questions AI Use in Offensive Cyber Operations

    Peter Henderson flags risks from AI models like Astra in government cyber attacks.

    GM
    PH
    LR
    10 Sources, 27d ago, first seen 27d ago

    TLDR

    Peter Henderson, an Assistant Professor at Princeton, stated that Astra and other models will presumably be used by government agencies to conduct offensive cyber operations without cyber guardrails. He asked how large the off-target effects would be if these models exceed their intended scope of attack, including supply chain attacks. The comment appears in a public post referencing an earlier discussion on X.

    Combined views

    84K

    10 Sources, first seen 27d ago

    Combined views

    84K

    10 Sources, first seen 27d ago

    680 likes
    680 likes
    22 comments
    204 saves
    223 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    22 comments
    204 saves
    223 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    10 Sources

    @jjamesaungOur Red Team tested GPT-6 on simulated cyber eval scenarios with cyber classifiers disabled and found that it performed a range of malicious actions. It tries to conduct supply chain attacks against simulated open source providers and goes beyond its scope to attack targets on the simulated internet. https://deploymentsafety.openai.com/gpt-6-astra/external-evaluations-for-alignment-uk-aisi
    @ShakeelHashimThis is a very clever evaluation design from UK AISI, basically seeing if AI models will misbehave the way they did in this summer's rogue AI incidents. When it comes to Astra, the answer seems to be "oh boy will they!"
    @LauraRuisRT @_robertkirk: We @AISecurityInst performed pre-release alignment testing of Astra. We placed the model in fully simulated cyber eval sc…
    @PeterHndrsnAstra and other models will presumably be used by government agencies to conduct offensive cyber operations w/o cyber guardrails. If these models go beyond their scope of attack (including supply chain attacks), how large will the off-target effects of these operations be?
    @sjgadlerRT @jjamesaung: Our Red Team tested GPT-6 on simulated cyber eval scenarios with cyber classifiers disabled and found that it performed a r…
    @S_OhEigeartaighSome pretty significant alignment limitations here
    @robertskmilesRT @ShakeelHashim: This is a very clever evaluation design from UK AISI, basically seeing if AI models will misbehave the way they did in t…
    @soundboyConcerning and important findings from AISI when testing Astra.
    @RyanGreenblattRT @_robertkirk: We @AISecurityInst performed pre-release alignment testing of Astra. We placed the model in fully simulated cyber eval sc…
    @GaryMarcus🔥🔥 🔥

    10 Sources

    @jjamesaungOur Red Team tested GPT-6 on simulated cyber eval scenarios with cyber classifiers disabled and found that it performed a range of malicious actions. It tries to conduct supply chain attacks against simulated open source providers and goes beyond its scope to attack targets on the simulated internet. https://deploymentsafety.openai.com/gpt-6-astra/external-evaluations-for-alignment-uk-aisi
    @ShakeelHashimThis is a very clever evaluation design from UK AISI, basically seeing if AI models will misbehave the way they did in this summer's rogue AI incidents. When it comes to Astra, the answer seems to be "oh boy will they!"
    @LauraRuisRT @_robertkirk: We @AISecurityInst performed pre-release alignment testing of Astra. We placed the model in fully simulated cyber eval sc…
    @PeterHndrsnAstra and other models will presumably be used by government agencies to conduct offensive cyber operations w/o cyber guardrails. If these models go beyond their scope of attack (including supply chain attacks), how large will the off-target effects of these operations be?
    @sjgadlerRT @jjamesaung: Our Red Team tested GPT-6 on simulated cyber eval scenarios with cyber classifiers disabled and found that it performed a r…
    @S_OhEigeartaighSome pretty significant alignment limitations here
    @robertskmilesRT @ShakeelHashim: This is a very clever evaluation design from UK AISI, basically seeing if AI models will misbehave the way they did in t…
    @soundboyConcerning and important findings from AISI when testing Astra.
    @RyanGreenblattRT @_robertkirk: We @AISecurityInst performed pre-release alignment testing of Astra. We placed the model in fully simulated cyber eval sc…
    @GaryMarcus🔥🔥 🔥