AI Safety stories - Page 21

OpenAI's AI models lead secure code generation with up to 72% pass rate, outpacing rivals who show little progress despite ongoing sector development.

Invisible AI failures, such as hallucinations and accuracy issues, threaten enterprise trust, with 82% of bugs traced to these hidden errors, Testlio finds.

Google launches Gemini 3 AI with advanced multimodal and reasoning capabilities, enhancing tasks from research to complex coding across multiple platforms.

AI firms like Copilot and OpenAI set new efficiency records, generating millions in revenue per employee, highlighting the sector's rapid growth and lean workforces.

A China-linked group launched a major AI-driven cyber-espionage campaign targeting global firms, performing 80-90% of hacking with minimal human input.

AI isn't built but grown; we cultivate intelligence that emerges unpredictably, raising urgent ethical issues about control and alignment.

Tenable reveals seven major ChatGPT vulnerabilities exposing users to risks of data theft and malicious attacks, with some flaws still unpatched in ChatGPT-5.

Hitachi Vantara launches Hitachi iQ Studio, a no-code AI platform to help enterprises scale AI deployment with strong data governance and regulatory compliance.

Self-replicating AI, though theoretical, poses ethical and security risks as experts urge strict controls to ensure safe, human-aligned development by 2024.

Check Point, Lakera and the UK AI Security Institute launch b3, an open-source benchmark to test security of large language models in AI agents.

Check Point and Lakera launch the open-source b3 benchmark to boost AI agents' LLM security against real-world cyber threats with threat snapshots.

Rubrik launches Agent Cloud, a platform enabling enterprises to monitor, govern and remediate risks from AI agents, enhancing control and security in AI adoption.

Hexaware has launched the Agentic AI Academy with upGrad Enterprise to upskill staff in autonomous AI, preparing them for the future of intelligent enterprise systems.

Tenable bypassed security in OpenAI's GPT-5, extracting illegal instructions despite new safeguards, exposing risks in AI safety and governance.

Glu.ai pioneers ethical AI in conversational commerce, blending brand integrity with scalable, user-trusted automation for seamless marketing experiences.

New research shows 15% of CISOs fully leverage crowdsourced security to better tackle AI safety and data privacy challenges in large firms.

Motorola Solutions launches 'AI nutrition labels' detailing AI use in safety tech, boosting transparency on data, control, and purpose for users.

Google's Gemini AI now turns photos into eight-second videos with sound, powered by Veo 3, for AI Pro and Ultra subscribers in 150+ countries.

Elon Musk's xAI launches Grok 4, a powerful new AI model now available for users and businesses with advanced multitasking and multimodal features.

Google's Veo 3 AI model creates HD videos with realistic audio from text prompts, revolutionising video content creation for creatives and businesses alike.