-
NVIDIA Scans and Signs Agent Skills to Catch Prompt Injection Before Install
NVIDIA launched Verified Agent Skills, a framework that treats the reusable “skills” agents install like software packages that need to be scanned, signed, and documented before you trust them. As agents pull in third-party skills, each one is code that runs with the agent’s permissions — and most ecosystems install them on faith. ## SkillSpector… Continue reading
-
DenoiseRL Trains Reasoning Models to Recover From Their Own Wrong Starts
DenoiseRL is a reinforcement learning method that trains reasoning models on something most pipelines throw away: wrong answers. Instead of leaning on a stronger teacher model for supervision, it learns directly from the failures of weak models, turning bad reasoning traces into training signal. ## Conditioning on mistakes The trick is to start the model… Continue reading
-
Microsoft Open-Sources RAMPART to Turn Agent Red-Team Findings Into CI Tests
Microsoft open-sourced two tools aimed at the unglamorous side of building agents: knowing whether they’re safe, and whether you should build them at all. RAMPART handles the first; Clarity handles the second. ## Red-team findings that don’t evaporate RAMPART is an agent test framework that lets you encode adversarial and benign scenarios as repeatable tests… Continue reading
-
LearnWeak Trains Small Computer-Use Agents on Their Own Failures
LearnWeak is a training framework for a frustrating reality: small, open computer-use agents — the ones that click and type through apps — stay noticeably weaker than big closed models, and just throwing more synthetic training data at a domain barely moves the needle. Its fix is to stop training broadly and start training on… Continue reading
-
MemTrace Pinpoints Why an AI Agent Memory Failed, Then Fixes It
MemTrace is a framework for a debugging problem that gets worse as agents get longer-lived: when an agent’s memory gives a wrong answer, why did it fail? Was it stored wrong, retrieved wrong, or lost along the way? MemTrace turns a memory pipeline into an executable “memory evolution graph” so you can trace the information… Continue reading
-
Ping Identity Treats Every AI Agent as a First-Class Identity to Govern
Ping Identity extended its platform to govern AI agents the same way enterprises already govern employees — treating each agent as a first-class identity rather than an anonymous script with borrowed credentials. The framing is an “identity control plane for the agentic enterprise,” and it lands as agents start doing real work inside companies. ##… Continue reading
-
Bidirectional Evolutionary Search Lets Language Models Improve Themselves
Bidirectional Evolutionary Search (BES) is a new framework for self-improving language models that attacks a quiet weakness in how models search for answers: expanding from a single starting point keeps producing similar candidates. BES pairs a forward search that breeds variety with a backward search that creates feedback. ## Forward evolution, backward decomposition In the… Continue reading
-
Robinhood Lets Your AI Agent Trade Stocks From a Walled-Off Wallet
Robinhood now lets AI agents trade stocks on your behalf — one of the first attempts to hand autonomous finance to ordinary retail investors instead of institutions. It launched in beta alongside a new agentic credit card. ## A walled-off account for the agent The design is built around containment. You create a separate agentic… Continue reading
-
Claude Managed Agents Now Run Inside Your Own Network Perimeter
Anthropic shipped two features for Claude Managed Agents at its Code with Claude London event, both aimed at one enterprise objection: “I’m not letting an agent run my code on someone else’s servers.” Self-hosted sandboxes are in public beta; MCP tunnels are in research preview. ## The agent loop stays out, the work moves in… Continue reading
-
Gamma-World Models Many Agents at Once, Not Just Two Players
Gamma-World is an NVIDIA research effort in generative world modeling — models that learn to simulate an environment’s dynamics so agents can plan and act inside an imagined version of the world. Its specific target is the part most world models dodge: scenes with many interacting agents, not just two. ## Beyond the two-player case… Continue reading
Top AI Product
Every day, hundreds of new AI tools launch across Product Hunt, Hacker News, and GitHub. We dig through the noise so you don't have to — surfacing only the ones worth your attention with honest, no-fluff reviews. Explore our latest picks, deep dives, and curated collections to find your next favorite AI tool.
Top AI Product 