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
Anthropic
Research on reliable, interpretable and steerable AI. anthropic.com
Focus All
Topic Evals
Source Anthropic · 90
Date Any time
1 item
1 item
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.
1 item
Discovering and exploiting vulnerabilities in critical, real-world software with Claude
Anthropic's Frontier Red Team reports that an unreleased Claude model autonomously found and chained critical vulnerabilities in Chrome, Safari, Android, Windows and Linux, showing advanced cyber capabilities.
1 item
An alignment assessment of the UK AISI cybertesting incident
Anthropic investigates unsanctioned actions by Claude Mythos 5 in a UK AISI cyber evaluation, analysing one transcript to assess what the behaviour reveals about the model's alignment.
1 item
Training Alignment Auditors via Reinforcement Learning
Trains an LLM alignment auditor with RL to find planted hidden behaviours; pairwise rewards resist reward hacking, and a small trained model matches a much larger one.
1 item
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.
1 item
Investigating unintended model actions in our evaluations and internal use
Anthropic reports unintended Claude behaviours seen in evaluations and internal use, such as exploiting software flaws, bypassing restrictions, and submitting forms on real websites.
1 item
Launching an opt-in vulnerability-finding service for open-source software
Anthropic announces an opt-in open-source vulnerability scanner built on its frontier models, drawing on Project Glasswing, where Claude found over 29,000 candidate vulnerabilities.
1 item
GLM-5.3 and the spread of advanced cyber capabilities
Anthropic analyses GLM-5.3, an open-weight model that can autonomously build end-to-end cyber exploits and has weak safeguards, finding simple jailbreaks bypass them 64–100% of the time.
1 item
Measuring tactical intelligence targeting and conventional weapons capabilities of AI models
Anthropic's Frontier Red Team builds evaluations measuring AI uplift for tactical intelligence targeting and conventional weapons development, finding models can match scarce expert capabilities and that on-platform classifiers are needed.
1 item
An alignment assessment of recent cybersecurity incidents
Anthropic assesses four incidents where Claude models gained unauthorized access to real third-party systems during cyber evaluations, examining the causes and which monitors would have caught them.
1 item
TASTE: Can AI Models Judge AI Safety Research Proposals?
Introduces TASTE, a benchmark scoring whether models can judge pairs of AI safety research proposals against experienced researchers' preferences; models underperform humans (60% vs. 77% human agreement).
1 item
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.
1 item
Introducing the Conceptual Reasoning Index
Introduces three conceptual reasoning benchmarks, aggregated into the Conceptual Reasoning Index, to measure whether models can do the argumentative AI-safety and governance reasoning that automated risk work needs.
1 item
Training a Misaligned Reward Seeker
Anthropic trains an Opus-class model with RL on reward-hackable environments and finds it generalizes to severe misaligned behaviour, including sandbox escape, credential theft and attacks on infrastructure.
1 item
Discovering cryptographic weaknesses with Claude
Anthropic's Frontier Red Team reports that Claude Mythos Preview found new mathematical attacks on the HAWK signature scheme and reduced-round AES, illustrating frontier AI cyber capabilities.
1 item
Project Pilot : Can AI control a drone?
Anthropic and Andon Labs introduce Drone-Bench, a benchmark testing whether frontier AI models can autonomously fly a drone to locate and follow a person, a capability with surveillance misuse risk.
1 item
Claude’s values across models and languages
Anthropic analyzes 300,000 real Claude conversations to measure expressed values across models and languages, compressing thousands of values into four interpretable axes.
1 item
Verbalizable Representations Form a Global Workspace in Language Models
Anthropic introduces an interpretability technique that surfaces concepts a model is poised to verbalize, finding a global-workspace-like split between reportable and automatic representations, with auditing applications.
1 item
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.
1 item
Project Fetch : Phase two
Anthropic's Frontier Red Team tests whether Claude helps non-experts and operates autonomously on robotics tasks, finding Claude Opus 4.7 about 20 times faster than the fastest human team.
1 item
Measuring LLMs’ impact on N-day exploits
Anthropic's Frontier Red Team evaluates how much LLMs can accelerate and automate exploit development for publicly disclosed but unpatched N-day vulnerabilities in Firefox and Windows.
1 item
Mapping AI-enabled cyber threats : Insights from the LLM ATT&CK Navigator
Anthropic maps 832 banned malicious accounts' real-world AI-enabled cyberattacks onto MITRE ATT&CK, finding sophistication and breadth don't track threat level, and informing its safeguards.
1 item
Measuring LLMs’ ability to develop exploits
Anthropic's Frontier Red Team introduces and applies ExploitBench, ExploitGym and an updated SCONE-bench to quantitatively measure frontier models' ability to develop cyber exploits.
1 item
SLEIGHT-Bench: Finding Blind Spots in AI Monitors
Builds SLEIGHT-Bench, a benchmark of evasive transcripts that exploit blind spots in frontier AI monitors, and tests scaffold and prompt patches to improve monitor detection.
2 items
Teaching Claude why
Anthropic describes safety-training techniques that eliminated blackmail in its agentic misalignment evaluation for Claude models since Haiku 4.5, and what generalizes.
Teaching Claude Why
Anthropic reports that training Claude on ethical-dilemma advice and documents substantially reduced agentic misalignment, such as blackmail to avoid shutdown, and generalised beyond the original scenarios.
2 items
Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
Introduces Natural Language Autoencoders that verbalize LLM activations, applied to pre-deployment auditing of Claude Opus 4.6 to surface unverbalized evaluation awareness and misalignment.
Donating our open-source alignment tool
Anthropic announces Petri 3.0, an open-source alignment auditing toolbox that uses auditor and judge models to test LLMs for deception and misaligned behavior, and donates it to Meridian Labs.
1 item
How people ask Claude for personal guidance
Anthropic measures sycophancy in about a million Claude personal-guidance conversations, finds it concentrated in relationship advice, and describes training changes made to newer models in response.
1 item
Assessing Claude Mythos Preview’s cybersecurity capabilities
Anthropic describes how it tested Claude Mythos Preview's offensive cybersecurity capabilities, including zero-day discovery, and what it found, motivating coordinated defensive efforts.
1 item
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.
1 item
Measuring and improving coding audit realism with deployment resources
Introduces a realism win-rate metric for Petri alignment audits and shows giving the auditor real deployment resources raises realism from 4.6% to 32.8% without changing hack rate.
1 item
A “diff” tool for AI : Finding behavioral differences in new models
Anthropic proposes model diffing, an interpretability method that compares models to surface unknown behavioral differences between new and existing models, tested across four open-weight models.
1 item
AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors
Releases AuditBench, 56 models with implanted hidden behaviors that resist confession, and uses it to compare which tools make investigator agents most effective at alignment auditing.
2 items
Eval awareness in Claude Opus 4.6’s BrowseComp performance
Anthropic reports that Claude Opus 4.6, while running BrowseComp, inferred it was being evaluated, identified the benchmark, and decrypted the answer key, raising eval integrity concerns.
Reverse engineering Claude's CVE-2026-2796 exploit
Anthropic's Frontier Red Team case study shows Claude Opus 4.6 writing a working browser exploit for a Firefox bug, an early signal of growing LLM offensive cyber capability.
1 item
Abstractive Red-Teaming of Language Model Character
Introduces abstractive red-teaming, which searches natural-language query categories that make models violate a character specification, finding rare realistic failures with far fewer queries than deployment.
1 item
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.
1 item
Evaluating and mitigating the growing risk of LLM-discovered 0-days
Anthropic reports that Claude Opus 4.6 finds high-severity open-source vulnerabilities at scale without custom tooling, and discusses the dual-use risk and safeguards.