AI Safety stories - Page 11

Google is rolling out Nano Banana 2 across Gemini, Search, Ads and Cloud, making its faster image model the default in key AI products.

F5 Labs launches monthly AI security leaderboards, ranking popular models on new indices of risk, resilience and cost under live attack.

A new NE2NE study finds over half of common AI benchmarks are “contaminated”, casting doubt on headline model rankings and marketing claims.

AI has become firms' top data security fear as identity and deepfake threats surge, exposing gaps in visibility, encryption and governance.

A rapid surge in OpenClaw AI assistant use has left tens of thousands of exposed systems and a trail of hijacked tools and malicious add-ons.

Confluent upgrades its AI toolkit to link agents over live data streams and add multivariate anomaly detection for faster issue spotting.

TELUS Digital warns that common AI persona prompts can quietly reshape models' moral judgements, undermining consistency in high‑risk use.

Security bosses worldwide are stalling agentic AI roll-outs, citing severe cyber risks and weak identity and access controls.

AI has overtaken all other threats as the top global data security risk, with firms warning its rapid spread magnifies existing vulnerabilities.

AI-powered cyberattacks are rising sharply, leaving Canadian businesses exposed as legacy systems meet rapid AI adoption and automated threats.

Red Hat launches AI Enterprise and AI 3.3, deepening its Nvidia tie-up to target production-grade, hybrid cloud agentic AI at scale.

CompTIA unveils AI Help Desk Essentials course to train frontline IT support teams in safe, effective use of generative AI chatbots.

Red Hat launches AI Enterprise “metal-to-agent” platform and AI 3.3 update to standardise governed AI operations across hybrid clouds.

VAST Data unveils PolicyEngine and TuningEngine to tighten AI agent governance and continuously fine-tune models for safer deployments.

OpenAI has hired veteran tech and people leader Arvind KC as Chief People Officer to steer hiring and workforce strategy as AI reshapes work.

Anthropic alleges Chinese labs DeepSeek, Moonshot and MiniMax ran vast illicit campaigns to copy Claude and bypass US export curbs.

Process intelligence and composable architecture are emerging as the missing ingredients to turn fragile AI pilots into resilient, traceable value.

Insider risk now costs organisations an average USD $19.5 million a year as AI reshapes data access and employee negligence drives losses.

One million professionals across 107 countries now use Thomson Reuters' CoCounsel AI, as the tool spreads through tightly regulated sectors.

Google's Gemini app can now spin 30-second AI songs from text or images using its new Lyria 3 model, with built-in SynthID watermarks.