← Back to People
Contributed to research evaluating AI models' ability to exploit blockchain smart contracts and discover zero-day vulnerabilities.
4
Mentions
2
Articles
1
Outlets
Oct 2025 — Dec 2025
Coverage
Associated Topics
Companies Covered
Coverage Patterns
How media typically covers Henry Sleight
Article Types
- research2
Media Angles
- safety1
- technical1
Narrative Framing
- cautionary1
Associated AI Models
Claude Opus 4.51Claude Sonnet 4.51GPT-51Claude1
Writes About
Appears Alongside
Nicholas CarliniResearcher
2 shared articlesWinnie XiaoResearcher
2 shared articlesAlwin PengResearcher
2 shared articlesCole KillianResearcher
2 shared articlesAlan ChanResearcher
2 shared articlesJifan ZhangResearcher
2 shared articlesAndi PengResearcher
2 shared articlesJohn SchulmanResearcher
2 shared articlesEsin DurmusResearcher
2 shared articlesArticles
Most recent firstArticles Written
Henry Sleight as author
Stress-Testing Model Specs Reveals Character Differences Among Language Models
AnthropicresearchneutralOct 27, 2025
Over 300,000 stress-test scenarios reveal that frontier language models from Anthropic, OpenAI, Google DeepMind, and xAI prioritize values differently and expose thousands of hidden contradictions and ambiguities in model specifications.
“Author of "Stress-Testing Model Specs Reveals Character Differences Among Language Models" in Anthropic”
AI AlignmentFoundation ModelsAI SafetyLLMs
Also Mentioned In
Referenced in coverage
In red teaming research from Anthropic, AI models found $4.6M worth of exploits in blockchain smart contracts
researchcautiousDec 4, 2025
“Co-authored research on AI models exploiting smart contracts and finding zero-day vulnerabilities.”
AI SafetyAI CybersecurityGenerative AIAI Research