✕
  1. X
  2. Gabriel Huang
Log inSign up
Gabriel Huang
129 posts
Gabriel Huang profile banner
user avatar

Gabriel Huang

@GabrielHuang9
Research Scientist @ServiceNow Research Agentic AI, LLMs, Security Testing, Red-Teaming, RLHF, Web Agents, Computer-Use Agents
Montréal, Québec
gabrielhuang.github.io
Joined March 2018
354
Following
419
Followers
RepliesRepliesMediaMedia

Log in or sign up for X

See what’s happening and join the conversation

Continue with phone
or
Log in with username or email
Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
  • Pinned
    user avatar
    Gabriel Huang
    @GabrielHuang9
    Jun 3
    We discovered that a small, open-weight AI model — freely available online — is sufficient to power a computer worm that autonomously adapts its attack to each machine it encounters. We built and tested one in an isolated lab to understand the threat. Conducted in an isolated
    Image
    00:00
  • user avatar
    Gabriel Huang
    @GabrielHuang9
    Jun 1
    Amazing work on LLM backdoors led by my colleagues @LeoBoisvert and @AbhayPuri98 at @ServiceNowRSRCH. if you've been training models in synthetic teacher traces in the open web, be careful because poisoned data/prompt injections in the environment can definitely propagate all the
    user avatar
    Abhay Puri
    @AbhayPuri98
    Jun 1
    Really happy to share that our paper -“Malice in Agentland: Down the Rabbit Hole of Backdoors in the AI Supply Chain” has won Best Paper (Outstanding Problem Paper) at @CAISconf 🏆 TL;DR - Poison ~2% of an AI agent’s fine-tuning traces and you can plant a trigger-activated
    Image
  • user avatar
    Gabriel Huang
    @GabrielHuang9
    Apr 23
    🚨 Looking for an intern at @ServiceNowRSRCH (Montréal) - composition, not capability. AI-native workflows are everywhere in demos, but most personal stacks quietly need a ton of human glue to stay useful. "Shadow agents" are now colliding with enterprise automations over shared
    Image
  • user avatar
    Gabriel Huang
    @GabrielHuang9
    Mar 24
    LiteLLM 18.2.8 steals api keys! It will run even without `import litellm`. Thanks to my colleagues for pointing it out to me
    user avatar
    Daniel Hnyk
    @hnykda
    Mar 24
    LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below
  • user avatar
    Gabriel Huang
    @GabrielHuang9
    Feb 28
    Props to anthropic for refusing to do mass surveillance and fully autonomous lethal weapons for the Pentagon. If a machine wants to kill someone, at a very minimum a human should have to approve it. anthropic.com/news/statement…
    user avatar
    Ilya Sutskever
    @ilyasut
    Feb 27
    It’s extremely good that Anthropic has not backed down, and it’s siginficant that OpenAI has taken a similar stance. In the future, there will be much more challenging situations of this nature, and it will be critical for the relevant leaders to rise up to the occasion, for