News
AI, cybersecurity, machine learning, scraping & programming news.
Are We Recovering Mechanisms? Objective-Level Recovery Gaps in Mechanistic Interpretability
Mechanistic interpretability seeks to recover the internal computations that produce model behavior. Circuit analysis represents a computation as a sparse subgraph of interacting model components. Detailed studies have reconstruct...
LLM2Jev: LLMs Are Already Jev-Style Decision Models -- When and How to Fine-Tune Them
Standard chat models generate free-form text that must be parsed into structured values, introducing latency and the risk of malformed outputs. Jev-style decision models return categorical probability distributions over predefined...
Homomorphic Advantage Operator: Stabilizing Reinforcement Learning Under Fully Homomorphic Encryption Constraints
Privacy-preserving machine learning presents significant deployment challenges on the cloud for intelligent systems with confidential data. Fully Homomorphic Encryption (FHE) offers a compelling solution for secure computation, pr...
Mass exploitation of Citrix NetScaler: What we currently know
Government agencies and critical infrastructure providers fell victim to a wave of attacks targeting Citrix NetScaler flaws that have been active for over a month. Security teams received direct warnings from government agencies a...
Show HN: Grist, open source coding harness
Government authorities and security teams are racing to assess the fallout from a campaign aimed at critical flaws in Citrix NetScaler. Government agencies and critical infrastructure providers were targeted in a wave of attacks d...
LibDragon (N64): Explicit Anti-AI Contributor Stance (GitHub Issue)
DragonMinded's libdragon project has opened an issue proposing an explicit ban on AI-generated contributions, joining a growing number of open source maintainers drawing hard lines against large language model output in their code...
Show HN: PhreshOS – OS for Web Apps
PhreshOS bills itself as an operating system purpose-built for software constructed with web technologies. The project positions itself as a runtime that lives on the machine it is installed upon, maintaining programs in execution...
Aweb – Communication for AI Agents
aweb provides durable messaging and identity management for AI agents, positioning itself as a communication layer that outlives individual sessions and spans machines, runtimes, and organizations. Messages are stored as server st...
Scan your LLM agent's tools for cost-amplification bugs (no API key)
Samuelrojas-dev's agent-cost-bench project provides a reproducible denial-of-wallet (DoW) benchmark for tool-using large language model agents, scoring outcomes on provider-billed tokens and USD rather than LLM judgments. The syst...
Kevin Mandia’s new ‘agent swarm’ security startup Armadin raises $255.5M at $2.5B valuation
Kevin Mandia, the founder who built Mandiant into a $5.4 billion acquisition by Google in 2022, has secured $255.5 million in Series B funding for his new venture Armadin at a valuation exceeding $2.5 billion. The round, announced...