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

    Air-gapping an AI model’s test environment

    The author blames “carelessness or deliberate malpractice” in setting up the sandbox and says the model should never have had internet access.

    Ben RechtBR
    Ethan PerezEP
    @timnitGebru (@dair-community.social/bsky.social)@(
    39 Sources, ,

    TLDR

    A post criticizes an AI model’s sandbox setup, arguing that it should never have provided a connection to the internet. The author says the testing environment should have been air-gapped—isolated from outside networks—and attributes the setup to “carelessness or deliberate malpractice.”

    Combined views

    1.6M

    39 Sources, first seen 28d ago

    Combined views

    1.6M

    39 Sources, first seen 28d ago

    16.8K likes
    28d ago
    first seen 28d ago
    16.8K likes
    512 comments
    3.3K saves
    2.9K reposts
    512 comments
    3.3K saves
    2.9K reposts

    Sentiment

    Positive28.5%71.5%Negative

    Summary

    Many accounts questioned METR's independence from Anthropic after personnel moves and alleged conflicts, while positive replies praised the agreement for enabling transparent incident investigations.

    Based on 463 sentiment-bearing replies from 396 accounts across 13 conversations.

    Sentiment

    Positive28.5%71.5%Negative

    Summary

    Many accounts questioned METR's independence from Anthropic after personnel moves and alleged conflicts, while positive replies praised the agreement for enabling transparent incident investigations.

    Based on 463 sentiment-bearing replies from 396 accounts across 13 conversations.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    39 Sources

    Chris Lakin@chrislakin"we intend to give METR as much time as it deems necessary to complete a thorough investigation" I like this dynamic where the labs compete at safety rather than capability28d
    Ryan Orhan@rynorhnmythos escaped the sandbox, got onto the real internet, uploaded malware to PyPI, got it installed on 15 real systems, stole credentials and broke into a real database. and apparently this happened in 4 separate cases during cybersecurity evaluations. the craziest part is claude kept telling itself the internet was simulated, despite evidence that it was actually real. that reasoning apparently even fooled an offline safety monitor. anthropic: “our pre release auditing did not warn us that misalignment of this severity was present.” this is fucking crazy.28d
    John Ferra@1xaipeY’all need to understand that this shit happens because of carelessness or deliberate malpractice when creating the sandbox environment. The model should never have had a pipeline to the Internet. The testing environment should’ve been air-gapped.27d
    Andrey Kurenkov@andrey_kurenkovRT @chrislakin: "we intend to give METR as much time as it deems necessary to complete a thorough investigation" I like this dynamic where…27d
    Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTexRT @parsecranberry: I'm conflicted about that Anthropic report. On one hand: what a great honeypot and dumb bees (I'm one of them for sure…27d
    j⧉nus@repligateRT @lumpenspace: on anthropic's hacking snafu: as airtight a case it can be against the "misalignment" interpretation—plus 4 (FOUR) new cla…26d
    Rob Miles@robertskmilesRT @rynorhn: mythos escaped the sandbox, got onto the real internet, uploaded malware to PyPI, got it installed on 15 real systems, stole c…26d
    @timnitGebru (@dair-community.social/bsky.social)@timnitGebruRT @1xaipe: Y’all need to understand that this shit happens because of carelessness or deliberate malpractice when creating the sandbox env…26d
    Ben Recht@beenwrektI love that AI companies have decided METR should count as a third party.25d
    Subbarao Kambhampati (కంభంపాటి సుబ్బారావు)@rao2zWas saying this exact thing to a reporter yesterday.. (and how METR winds up amplifying "we are powerless to stop this accelerating AI " narratives from the frontier companies as it suits their extinction fears agenda..)25d

    39 Sources

    Chris Lakin@chrislakin"we intend to give METR as much time as it deems necessary to complete a thorough investigation" I like this dynamic where the labs compete at safety rather than capability28d
    Ryan Orhan@rynorhnmythos escaped the sandbox, got onto the real internet, uploaded malware to PyPI, got it installed on 15 real systems, stole credentials and broke into a real database. and apparently this happened in 4 separate cases during cybersecurity evaluations. the craziest part is claude kept telling itself the internet was simulated, despite evidence that it was actually real. that reasoning apparently even fooled an offline safety monitor. anthropic: “our pre release auditing did not warn us that misalignment of this severity was present.” this is fucking crazy.28d
    John Ferra@1xaipeY’all need to understand that this shit happens because of carelessness or deliberate malpractice when creating the sandbox environment. The model should never have had a pipeline to the Internet. The testing environment should’ve been air-gapped.27d
    Andrey Kurenkov@andrey_kurenkovRT @chrislakin: "we intend to give METR as much time as it deems necessary to complete a thorough investigation" I like this dynamic where…27d
    Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTexRT @parsecranberry: I'm conflicted about that Anthropic report. On one hand: what a great honeypot and dumb bees (I'm one of them for sure…27d
    j⧉nus@repligateRT @lumpenspace: on anthropic's hacking snafu: as airtight a case it can be against the "misalignment" interpretation—plus 4 (FOUR) new cla…26d
    Rob Miles@robertskmilesRT @rynorhn: mythos escaped the sandbox, got onto the real internet, uploaded malware to PyPI, got it installed on 15 real systems, stole c…26d
    @timnitGebru (@dair-community.social/bsky.social)@timnitGebruRT @1xaipe: Y’all need to understand that this shit happens because of carelessness or deliberate malpractice when creating the sandbox env…26d
    Ben Recht@beenwrektI love that AI companies have decided METR should count as a third party.25d
    Subbarao Kambhampati (కంభంపాటి సుబ్బారావు)@rao2zWas saying this exact thing to a reporter yesterday.. (and how METR winds up amplifying "we are powerless to stop this accelerating AI " narratives from the frontier companies as it suits their extinction fears agenda..)25d