Ben Jennings on the Bayeux tapestry – cartoon23m ago The latest Avatar: Seven Havens trailer gives us a deeper dive into the characters and story25m ago Agentic AI Is The New Attack Surface: How Can SAP Application Security Teams Repel It?26m ago Euno raises $23m to give enterprise AI agents the context they lack29m ago Red Hat AI 3.5 tackles the GPU queue that can stall AI pilots39m ago A Candid Abacus AI Review: The All-in-One AI Platform for Professionals & Enterprises40m ago Governments are turning to Claude to automate spying40m ago CNN Just Ran One of the Worst AI Puff Pieces We’ve Ever Seen in Our Lives About That “AI Actor” Tilly Northwood53m ago How Credit Genie keeps codebase docs fresh with OpenWiki54m ago Closing The Gap Between AI Ambition And Data Reality56m ago Universal Music is launching an AI music platform with ElevenLabs1h ago Top Pentagon Official Rejects Anthropic Researcher’s AI Warning2h ago Arlequin AI raises €28M to build novel AI models that learn complex relationships at scale1h ago SpaceX sues to block release of tax-break records for its Texas Terafab project1h ago

Ben Jennings on the Bayeux tapestry – cartoon23m ago The latest Avatar: Seven Havens trailer gives us a deeper dive into the characters and story25m ago Agentic AI Is The New Attack Surface: How Can SAP Application Security Teams Repel It?26m ago Euno raises $23m to give enterprise AI agents the context they lack29m ago Red Hat AI 3.5 tackles the GPU queue that can stall AI pilots39m ago A Candid Abacus AI Review: The All-in-One AI Platform for Professionals & Enterprises40m ago Governments are turning to Claude to automate spying40m ago CNN Just Ran One of the Worst AI Puff Pieces We’ve Ever Seen in Our Lives About That “AI Actor” Tilly Northwood53m ago How Credit Genie keeps codebase docs fresh with OpenWiki54m ago Closing The Gap Between AI Ambition And Data Reality56m ago Universal Music is launching an AI music platform with ElevenLabs1h ago Top Pentagon Official Rejects Anthropic Researcher’s AI Warning2h ago Arlequin AI raises €28M to build novel AI models that learn complex relationships at scale1h ago SpaceX sues to block release of tax-break records for its Texas Terafab project1h ago

MosaicLeaks: Can your research agent keep a secret?

Hugging FacePublished Jun 18, 2026

MosaicLeaks is a new tool designed to test whether AI research agents can maintain confidentiality when handling sensitive information. It evaluates agents' ability to keep secrets during interactions, highlighting potential privacy risks in AI research workflows.

#ai-research-agents #confidentiality #mosaicleaks #privacy

More in AI Research

AI Research4 min read

An Anthropic researcher just quit, saying OpenAI and Anthropic are 'gambling with our lives'

Covered by 12 sources1d ago

AI Research3 min read

Worried Anthropic researchers warn that AI ‘could kill all humans’

Covered by 8 sources1d ago

AI Research4 min read

Clay raises $115m at a $7.1bn valuation, more than double its 2025 price

The Next Web1h ago