A unified encoder for safety moderation and PII detection in a single forward pass, built on top of GLiNER2.đź“„ Paper · 🤗 ModelsGuardrails in an agentic LLM application5 classifiers Ă— 4 hops = 20 forward passes per requestSame application after GLiNER Gu…
Google caught attackers using AI to weaponize a real bug. The frontier labs weren’t the source. The lesson is uglier than it looks.Continue reading on Towards AI »
By Michal Florek, October 2025 (Updated May 2026)Executive SummaryIn the accelerating age of artificial intelligence, the world faces a subtle but existential tension: machines are becoming flawless, while humans are not.Yet it is precisely our flaws —…
AI Safety, Security and LegislationContinue reading on Medium »
The next AI safety problem is not whether chatbots are intelligent. It is whether they are intimate enough to be believed.Continue reading on Activated Thinker »
The real problem is not context length. It is self-generated context pollution.Continue reading on Medium »
What We Found Inside Should Change How You Think About Every Benchmark.Continue reading on Predict »
What We Found Inside Should Change How You Think About Every Benchmark.Continue reading on Predict »
AI Safety in PracticeOver the past few years, large language models (LLMs) have become embedded in many business operations, including powering customer service platforms, internal knowledge tools, and automated decision systems.But rapid adoption has …
Insider Brief A new paper from researchers affiliated with Google DeepMind, OpenAI, Anthropic and several universities reports that the artificial intelligence industry is too focused on preventing harm and not focused enough on helping humans thrive, outlining a framework called “positive alignment” that would train AI systems to actively support human flourishing rather than merely […]