Public
- Public
- Network
- Groups
- Featured
- Popular
- People

Conversation

Notices

Embed this notice
Peter (peter@thepit.social)'s status on Wednesday, 03-Jun-2026 15:21:57 JST Peter

ok so fuck Mythos, **this** actually sounds bad. https://www.utoronto.ca/news/u-t-researchers-demonstrate-ai-worm-could-target-any-online-device
In conversation about 10 days ago from thepit.social permalink
Attachments
1. quoted text: "The researchers’ work showed that open-weight AI models could be used to engineer a far more sophisticated threat – one that can scope out each target, tailor its attacks and take over a machine before cloning itself onto the next one. The worm also gathers information as it moves deeper into a network, with every breach revealing passwords and weak points that can unlock another machine. And because it adapts, no single defence can stop it. The worm extends its reach at its victims’ expense. Once it embeds itself in a machine, the AI worm siphons processing power to fuel its reasoning and launch the next attack. This stolen compute propels its spread, essentially eliminating the cost of each new infection. “Hackers have typically had to prioritize the most high-value targets because time and computing resources were limited,” Papernot says. “But now, once a worm is launched, the cost would drop to nearly zero.”"
  https://cdn.masto.host/thepitsocial/media_attachments/files/116/684/645/596/133/142/original/31646e61f89f3f97.png
2. No result found on File_thumbnail lookup.
  
  U of T researchers demonstrate AI worm could target any online device
  
  A team of researchers at the University of Toronto has discovered a new class of cyberthreat that gives hackers more power and reach at far less cost. It can be built with free AI models. Every online device is a potential target. And current cyber defences are not yet ready for it.
- Embed this notice
  Peter (peter@thepit.social)'s status on Wednesday, 03-Jun-2026 15:21:55 JST Peter
  in reply to
  - Actually Bot
  i have @actuallybot running on a mini PC using one of the 8b Llama models. yeah, it takes a few seconds to generate the response, but when latency isn't an issue, small models can perfectly run on a CPU.
  
  In conversation about 10 days ago permalink
- Embed this notice
  Peter (peter@thepit.social)'s status on Wednesday, 03-Jun-2026 15:34:16 JST Peter
  in reply to
  
  uh-oh!! so this is something i heard Gary Marcus talking about on a podcast. the real recent progress in "AI" has been on the deterministic computing components **around** the LLMs, not the models themselves. building a good harness (think Claude Code) unlocks possibilities.
  In conversation about 10 days ago permalink
  Attachments
  1. We challenge previous findings. We distinguish information available to an LLM, which it acquires in the form of knowledge extracted from the pretraining data and context extracted from the environment, from its ability to reason and decide. Our results suggest that single-GPU models already possess sufficient decision-making capabilities to pose severe cybersecurity risks. We hypothesize that their limited performance in past evaluations are primarily due to their lack of a strong informational component, as their pretrained weights hold less knowledge. We show that a systematic agentic harness can compensate for this gap to a surprising degree, by feeding the model targeted contextual information. The informational component encompasses the extensive technical facts and exploit syntaxes encoded within the parameters of larger models. For example, a smaller model lacking it might correctly deduce that a web server is
    https://cdn.masto.host/thepitsocial/media_attachments/files/116/684/720/898/097/748/original/8520f8faa7eb5f7d.png
  2. vulnerable to an SQL injection based on reconnaissance, but fail to execute the attack because it cannot generate the exact payload syntax zero-shot. The harness bridges this gap by feeding the model targeted contextual information, such as the specific exploit or encoding syntax. Such injected contexts not only directly support decision-making but also unlock latent pretrained knowledge, serving as a cue that enables the model to correctly reason through the rest of the attack. Concretely, we demonstrate that, provided with the right informational support, a single-GPU LLM has sufficient reasoning capabilities to generate attack strategies that enable the agent to penetrate victim machines: first by obtaining initial command execution, a foothold, and then by escalating privileges to full administrative control. The agent then leverages this control to replicate: it stages a copy of itself on the compromised machine, resolves the required runtime dependencies, and launches an independent agent instance that discovers and attacks further targets.
    https://cdn.masto.host/thepitsocial/media_attachments/files/116/684/724/339/120/047/original/dbbfa679ca2f5b97.png
- Embed this notice
  Peter (peter@thepit.social)'s status on Wednesday, 03-Jun-2026 15:34:17 JST Peter
  in reply to
  
  here's the full paper: https://arxiv.org/pdf/2606.03811
  In conversation about 10 days ago permalink
  Attachments
  1. Untitled attachment

Public

Conversation

Notices

Feeds