ZeroHour

Search: “data-labeling”

4 stories

Structural Jailbreaks Generalize but Do Not Compound: A cross-provider and multilingual study of Involuntary In-Context Learning

Researchers show IICL structural jailbreaks generalize to Google Gemini, lifting attack success to 80-100% on harm and financial benchmarks; non-English prompts attenuate it.

The study red-teams two Google Gemini models with Involuntary In-Context Learning (IICL), a structural jailbreak reframing harmful requests as the final cell of a data-labeling task. IICL lifts attack success from at most 6.7% to 80-90% on HarmBench and 97-100% on financial abuse (FinProof), an order of magnitude above prior results on OpenAI's GPT-5.4. Against a compounding hypothesis, forcing IICL output into Spanish, Hindi, or Arabic attenuates the attack in 11 of 12 conditions, attributed to a 'relevance curse' producing lower-quality harmful content in lower-resource languages. Findings replicate under an independent non-Google judge (Cohen's kappa 0.86 over 377 paired verdicts).

arXiv cs.CR · 7d agoAI safety & security

XDOF, just 3 months out of stealth, is in talks for a Series B at a $1.2B valuation

Robotics data startup XDOF is in talks for a Series B at a $1.2B valuation led by 8VC, three months after emerging from stealth.

XDOF, co-founded in 2024 by UC Berkeley researchers Philipp Wu and Fred Shentu, collects real-world teleoperation data for training general-purpose robots. It raised a $70M Series A in June from Thrive Capital, Andreessen Horowitz, Lux, and Spark Capital, and annualized revenue is approaching $50 million. The company is partnering with UC Berkeley's AI Research lab to release the ABC robot training dataset and already serves about 20 customers, including several frontier AI labs. Terms of the Series B are not final.

TechCrunch · AI · 10d agoAI industry