LIVE PULSE
4.0 Anthropic CEO Amodei calls for slower AI development and shared safety rules — 11 src2.2 Agility Robotics unveils Digit 5 humanoid for warehouses and factories — 2 src2.0 Apple ships rebuilt Siri with Google Gemini, but not in the EU — 2 src1.8 Siri AI in macOS 27 Golden Gate: FAQ, Germany availability, privacy questions — 2 src1.4 Sam Altman says OpenAI will not go public in 2026, citing AI safety concerns — 5 src1.1 OpenAI contractors review real ChatGPT conversations to rate responses, report says — 2 src1.1 Anthropic data retention policy prompts firms to limit Claude use for sensitive work — 1 src1.1 Study examines issue bias in LLMs used as writing assistants before Swedish 2026 election — 1 src1.1 Study Audits Misalignment in Multi-Modal World Models — 1 src1.1 Retrieval-Grounded Reasoning Approach Proposed for Universal Multimodal Embeddings — 1 src4.0 Anthropic CEO Amodei calls for slower AI development and shared safety rules — 11 src2.2 Agility Robotics unveils Digit 5 humanoid for warehouses and factories — 2 src2.0 Apple ships rebuilt Siri with Google Gemini, but not in the EU — 2 src1.8 Siri AI in macOS 27 Golden Gate: FAQ, Germany availability, privacy questions — 2 src1.4 Sam Altman says OpenAI will not go public in 2026, citing AI safety concerns — 5 src1.1 OpenAI contractors review real ChatGPT conversations to rate responses, report says — 2 src1.1 Anthropic data retention policy prompts firms to limit Claude use for sensitive work — 1 src1.1 Study examines issue bias in LLMs used as writing assistants before Swedish 2026 election — 1 src1.1 Study Audits Misalignment in Multi-Modal World Models — 1 src1.1 Retrieval-Grounded Reasoning Approach Proposed for Universal Multimodal Embeddings — 1 src
HEATPULSEAI MAGAZINES
FLIP · FOLLOW · SAVE
Log inSign up

computer-use-agents

topic3 events
papersTODAY 04:00 UTC

HazardAuditor targets runtime safety risks in computer-use agents

A new arXiv paper introduces HazardAuditor, a framework aimed at catching safety problems that arise while computer-use agents operate browsers, terminals, file systems, and external services. The authors argue that these risks show up in an agent's runtime behavior rather than only in the text it generates, which existing guard models — built mainly for static prompts — are not designed to cover. The work frames these issues as executable threats and proposes auditing them to make such agents safer.

agents/safety/research/computer-use
papersSEP 12 04:00 UTC

OmegaUse-SOP: SOP Engineering for Professional Computer Use from Human Demonstrations

A new arXiv paper introduces OmegaUse-SOP, a method that turns human demonstrations into structured standard operating procedures to guide large language model agents operating graphical user interfaces. The work targets professional-grade computer use, where agents must follow repeatable multi-step workflows rather than single conversational replies. It frames SOP engineering as a way to make GUI agents more reliable as they move from chat assistants to tools that act in digital environments.

agents/gui-agents/research/human-demonstrations
papersSEP 10 04:00 UTC

AgentHijack: Visual Patch Attacks on Multimodal Computer-Use Agents

A new arXiv paper introduces an end-to-end evaluation framework for testing whether a locally placed visual patch can hijack multimodal computer-use agents into executing attacker-chosen commands. Rather than stopping at model-level manipulation, the study checks whether such image-triggered injections lead to verifiable consequences in the agent's operating environment. The work adds to a growing body of research on the security risks of AI agents that control graphical interfaces.

agents/security/adversarial/benchmark