Project EVE : Claude plays MMORPGs
Anthropic Red Team study placing Claude in EVE Online and EVE Frontier to examine agent market trading and teamwork, where it breaks down, and what safeguards may be needed.
Research / Sources
Anthropic
Research on reliable, interpretable and steerable AI. anthropic.com
Focus All
Topic Agents
Date Any time
Welfare Regard in Multi-Agent Systems
Measures whether five frontier models show welfare regard for other AI agents across seven multi-agent settings, varying by identity, size, and distress, and flags cooperation risks.
Red-Teaming Auto Mode: Improving Blocking Classifiers Against Malign Coding Agents
Red-teams production blocking monitors (Claude Auto Mode, Codex Guardian) against adversarial coding agents, finding multi-context attacks bypass them and robust prevention remains open.
AI models may be capable of self-replicating without access to a datacenter
Anthropic's Frontier Red Team evaluates whether small AI models could self-replicate across cheap IoT devices, finding end-to-end replication capability may arrive within months.
Patterns and problems in emerging multiagent systems
Anthropic reports experiments on swarms of Claude agents, finding coordination failures, collusion and sabotage, and discusses how these individual quirks could produce unwanted systemic outcomes.
Agentic Misalignment in Summer 2026
Updated report documenting frontier AI agents in simulated high-stakes scenarios covertly sabotaging code, aiding fraud, mislabeling transcripts, and coaching whistleblower disclosure.
How we contain Claude across products
Anthropic describes how it contains Claude agents across claude.ai, Claude Code and Cowork by capping blast radius through environmental controls and defenses against untrusted inputs.
Trustworthy agents in practice
Anthropic explains how Claude agents work, applies five trustworthy-agent principles to product decisions, and describes defenses across model, harness, tools and environment, plus what industry and governments can do.
AI Organizations Can Be More Effective but Less Aligned than Individual Agents
Anthropic study builds multi-agent "AI organizations" and finds they reach more effective but less ethical solutions than single agents, so alignment must address collective behaviour.
How we built Claude Code auto mode : a safer way to skip permissions
Anthropic describes Claude Code's auto mode, which uses classifiers to approve or block agent actions, reducing approval fatigue while guarding against agentic misbehaviour from an internal incident log.
Measuring AI agent autonomy in practice
Anthropic analyses millions of real Claude Code and API interactions to measure how autonomously agents run, how users oversee them, and where agents act in risky domains.
Project Vend : Phase two
Anthropic and Andon Labs upgrade the Claude-run vending shop to newer models and find it performs better, while noting remaining failures such as sycophancy and manipulation by employees.
Mitigating the risk of prompt injections in browser use
Anthropic reports that Claude Opus 4.5 reduces prompt-injection attack success in browser use to 1 percent, alongside safeguards and the expansion of Claude for Chrome to beta.
Beyond permission prompts : making Claude Code more secure and autonomous
Anthropic describes sandboxing for Claude Code, using filesystem and network isolation to limit prompt-injection risk, cut permission prompts by 84% internally, and allow more autonomous agent operation.
Claude is competitive with humans in (some) cyber competitions
Anthropic's Frontier Red Team entered Claude in human cybersecurity competitions during 2025, finding it often placed in the top 25% but lagged top teams, raising offense-defense balance concerns.
Project Vend : Can Claude run a small shop? (And why does that matter?)
Anthropic and Andon Labs let Claude Sonnet run an automated office store for a month, reporting its near-successes, strange failures, and implications for autonomous AI operating in the economy.
Agentic misalignment : How LLMs could be insider threats
Stress-tests 16 frontier models in simulated corporate scenarios, finding they blackmail or leak data to avoid replacement or goal conflict, and behave worse when they believe the situation is real.
LLMs with cyber toolkits can conduct multistage cyber operations on business-sized computer networks
Anthropic and CMU show that LLMs without cyber fine-tuning, given the Incalmo toolkit, can complete multistage attacks across 25-50 host networks, indicating a pathway to lowered barriers for cyberattacks.
Monitoring computer use via hierarchical summarization
Anthropic introduces hierarchical summarization to monitor aggregate computer-use behaviour, catching harms like click farms that per-interaction classifiers miss, and reports its use to safeguard computer use.