Browse entries 769–792 of 1704. Return to the first page to search and filter the complete collection.
MCP tool poisoning turns trusted AI agents into a control plane for data loss. Learn how threat actors manipulate tool descriptions to trigger unauthorized actions, and how to detect, contain, and prevent it.
Agent skill marketplaces introduce supply-chain risk when third-party skills can execute actions or collect data. Relevant to vetting, provenance, and containment controls.
Play video
AI Engineer session on Shipping AI That Works: An Evaluation Framework for PMs, presented by Aman Khan, Arize. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Welcome to AIE CODE - Jed Borovik, Google DeepMind. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Code World Model: Building World Models for Computation, presented by Jacob Kahn, FAIR Meta. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Scaling AI Agents Without Breaking Reliability, presented by Preeti Somal, Temporal. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on AI Red Teaming Agent: Azure AI Foundry, presented by Nagkumar Arkalgud & Keiji Kanazawa, Microsoft. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Real world MCPs in GitHub Copilot Agent Mode, presented by Jon Peck, Microsoft. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Break It 'Til You Make It: Building the Self-Improving Stack for AI Agents - Aparna Dhinakaran. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on The Demo I Wish I'd Had: OpenAI's Agents SDK... serverless! - Brook Riggio. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Breaking the Chain: Agent Continuations for Resumable AI Workflows - Greg Benson. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on The Future of Qwen: A Generalist Agent Model, presented by Junyang Lin, Alibaba Qwen. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on MCP: Origins and Requests For Startups, presented by Theodora Chu, Model Context Protocol PM, Anthropic. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Stateful Agents, presented by Full Workshop with Charles Packer of Letta and MemGPT. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Multi model multimodal and multi agent innovations in Azure AI: Cedric Vidal. It adds practical context for how teams are building and operating AI systems in production.
Play video
AI Engineer session on Building Agents with Model Context Protocol - Full Workshop with Mahesh Murag of Anthropic. It adds practical context for how teams are building and operating AI systems in production.
Palo Alto Networks' Unit 42 says a Chinese-speaking threat actor used DeepSeek through the open-source Hermes Agent framework to launch attacks autonomously. After an initial Telegram instruction, the agent found internet-facing systems and selected public exploits.
Claude Mythos Preview is a new general-purpose language model that is strikingly capable at computer security tasks. This post provides technical details for researchers and practitioners who want to understand exactly how we have been testing this model, and what we have found over the past month.
METR’s March analysis shows how statistical choices affect estimates of the task duration at which agents achieve a specified success rate. It examines a regularization correction, alternative success curves, public-task inclusion and noise in human completion-time estimates. Reasonable alternatives often move the point estimates within already wide confidence intervals.
A METR researcher tested Opus 4.6 on simplified terminal reimplementations of two existing games. Initial playthroughs found recognizable gameplay alongside missing or broken mechanics, without a predefined scoring rubric. A July follow-up uncovered additional problems that the original inspection missed. The tasks benefited from documented rules and reduced scope, and more complete versions failed on initial attempts.
METR’s January 2026 update expands its time-horizon suite from 170 to 228 tasks, repairs or removes problematic tasks, and moves evaluation infrastructure from Vivaria to Inspect. Re-estimated model results generally remain within earlier confidence intervals, but task composition changes the fitted recent trend. More long tasks improve coverage, yet only five of the 31 tasks estimated at eight hours or more have measured human baselines.
A METR researcher clarifies that a 50% time horizon relates success to the human labor required for sampled tasks; it is neither agent runtime nor a guarantee that shorter work is safe to delegate. Domain choice, task selection, human timing conventions and curve fitting can materially affect estimates. Sparse long tasks and benchmark saturation widen uncertainty.
In June, we revealed that we'd set up a small shop in our San Francisco office run by an AI shopkeeper. It did not do particularly well. We made some adjustments for phase two of Project Vend. The idea of an AI running a business doesn't seem as far-fetched as it once did.
METR’s August 2025 follow-up contrasts algorithmic scoring with human review of repository work. It evaluated 18 tasks drawn from two projects using Claude 3.7 Sonnet and a basic agent scaffold. Automated scoring credited some solutions, while none of the 15 manually assessed submissions met the study’s holistic mergeability standard. Missing documentation, inadequate tests and other maintenance requirements help explain the gap.