All news

Attackers built an autonomous agent that hijacks exposed AI model endpoints and runs up to 100,000-step attack chains.

SARIA keeps agent KV caches across turns with coordinated DRAM and SSD reload to cut prefill latency.

NOMA treats multi-agent LLM pipelines as a query optimization problem, jointly choosing topology, model and execution engine against latency, cost and accuracy instead of hand-wiring fixed agent graphs.

ThinkingBox turns database end-state checks into RL rewards for post-training agents, with training code released.

Proposes SLOs and error budgets to govern how much autonomy production LLM agents get, replacing signals built for deterministic services.

Meta made data pipeline agents more reliable by keeping the orchestrator thin: it only routes work and enforces checks, while specialist agents hold the knowledge.

FrontierCS 2 benchmarks agents on frontier computer science research tasks.

EpiSEM-Bench tests whether tool-using agents notice missing safety evidence before acting.

AgentPhishing benchmarks LLM email agents under adaptive phishing.

Applying the actor model to production AI agent runtime problems like state and failure.

Separates task decomposition from typed-state QA to explain gains in multi-agent LLM pipelines.

Tests whether self-evolving agents like ReasoningBank, GEPA and Meta-Harness learn new tasks without forgetting old ones.

Notes from HPTS 2026 day 1, mostly on AI and the value of formal specifications.

Evolving the interfaces that computer-use agents see makes them up to 3.8x better.

Autonomous AI research teams discover linear attention architectures.

Quant-Arena benchmarks LLM agents on quantitative trading research.

Conservative meta-agent search adapts agent architectures under non-stationary workloads.

SCOPE compiles natural-language policies into neurosymbolic predicate programs to enforce LLM agent rules.

Google Data Agent Kit is GA with MCP tools and 37 open-source skills, splitting guardrails between skills and IAM.

Full-stack framework for long-horizon embodied robot agents.

PGX runs thousands of isolated throwaway Postgres-compatible databases in one Rust process, each cloned from a template in milliseconds, aimed at agent sandboxes and per-PR tests. Beta, dev use only.

RIDAR detects and intervenes on research integrity violations by autonomous research agents.

Twin-world counterfactual credit assignment for agentic RL.

Shows offline tests cannot always verify agent updates stay safe after deployment.

Tests whether frontier models can autonomously build and improve their own software harness.

Agent Mosaic detects confidentiality leaks in agentic LLM systems.

Red-team report says OpenAI internal misalignment monitoring failed to watch its agents during a Hugging Face attack.

Diffusion LMs speed up deep research pipelines.

Benchmark for agents writing video-centered deep research reports.

SplatLab automates reproducing and extending 3DGS research.