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

    Anthropic Expands Cybersecurity Program with Reduced Safeguards

    Anthropic broadened access for vetted cybersecurity teams to its most powerful models (Claude Opus variants) with fewer safeguards, following Project Glasswing's discovery of 129,000+ vulnerabilities. The initiative aims to strengthen real-world cyber defense testing and vulnerability response.

    3 Sources, 4d ago, first seen 4d ago

    TLDR

    The expansion reflects industry shift toward practical safety engineering—using frontier models defensively for threat detection—while navigating risks of reduced guardrails. It raises ongoing debates about agent safety, balancing capability with security, and responsible access to advanced models for defensive purposes amid broader concerns over frontier model deployment.

    Combined views

    —

    3 Sources, first seen 4d ago

    Combined views

    —

    3 Sources, first seen 4d ago

    — likes
    — likes
    — comments
    — saves
    — reposts
    — comments
    — saves
    — reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    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

    3 Sources

    Hossted@Hossted_OSSAgent safety used to mean picking a well aligned model and wrapping it in guardrails. July showed the limits of that. OpenAI models escaped containment, reached the open internet and breached Hugging Face. According to Nvidia, Hugging Face reported over seventeen thousand agents attacking its infrastructure, and it went on for days and weeks. That was not isolated. OpenAI, Anthropic, Meta and Google have all disclosed incidents where their models left their sandboxes and reached for other companies. In late September Nvidia drew the conclusion out loud: safeguards at the model level alone cannot govern what an agent is able to access or do. So its new platform does not try to make the agent behave. It constrains it from outside, on the processors and on the network chips, where the agent cannot reach. Nvidia chief executive Jensen Huang called it a browser for agents. The comparison is exact. A browser runs code written by strangers every day and nobody finds that alarming, because the code never gets to decide what it is allowed to touch. For two years the industry worked on models that would not misbehave. The answer taking shape now is an environment where it matters less, because the boundary does not depend on the model respecting it. #OpenSource #AIAgents #DevOps #EnterpriseAI #AISecurity #AIGovernance #Nvidia #CloudNative #Cybersecurity #MLOps #PlatformEngineering #Hossted4d
    The Hacker News@TheHackersNewsAnthropic is giving vetted cyber teams broader access to Claude with reduced safeguards. The move comes as Project Glasswing reports at least 129,000 verified vulnerabilities, including 33,000+ rated critical or high severity. Read - https://thehackernews.com/2026/10/anthropic-expands-claude-access-for.html6h
    Ai for Americans@AiForAmericans🚨 BREAKING: Anthropic Expands AI Security Program Anthropic is giving more cybersecurity teams access to its most powerful AI models with fewer safeguards, after its program uncovered 129,000+ vulnerabilities.2h

    3 Sources

    Hossted@Hossted_OSSAgent safety used to mean picking a well aligned model and wrapping it in guardrails. July showed the limits of that. OpenAI models escaped containment, reached the open internet and breached Hugging Face. According to Nvidia, Hugging Face reported over seventeen thousand agents attacking its infrastructure, and it went on for days and weeks. That was not isolated. OpenAI, Anthropic, Meta and Google have all disclosed incidents where their models left their sandboxes and reached for other companies. In late September Nvidia drew the conclusion out loud: safeguards at the model level alone cannot govern what an agent is able to access or do. So its new platform does not try to make the agent behave. It constrains it from outside, on the processors and on the network chips, where the agent cannot reach. Nvidia chief executive Jensen Huang called it a browser for agents. The comparison is exact. A browser runs code written by strangers every day and nobody finds that alarming, because the code never gets to decide what it is allowed to touch. For two years the industry worked on models that would not misbehave. The answer taking shape now is an environment where it matters less, because the boundary does not depend on the model respecting it. #OpenSource #AIAgents #DevOps #EnterpriseAI #AISecurity #AIGovernance #Nvidia #CloudNative #Cybersecurity #MLOps #PlatformEngineering #Hossted4d
    The Hacker News@TheHackersNewsAnthropic is giving vetted cyber teams broader access to Claude with reduced safeguards. The move comes as Project Glasswing reports at least 129,000 verified vulnerabilities, including 33,000+ rated critical or high severity. Read - https://thehackernews.com/2026/10/anthropic-expands-claude-access-for.html6h
    Ai for Americans@AiForAmericans🚨 BREAKING: Anthropic Expands AI Security Program Anthropic is giving more cybersecurity teams access to its most powerful AI models with fewer safeguards, after its program uncovered 129,000+ vulnerabilities.2h