AI News
Stay ahead with the most important AI news in one place.
The 2026 RSA Conference 2026 (RSAC), San Francisco has underscored a decisive shift in cybersecurity, with artificial intelligence now central to both defence and threat landscapes. A dominant theme is the rise of “agentic AI,” with industry commentary noting that “we are moving from copilots to fully autonomous agents” (Investors Business Daily).
Instagram is reportedly removing end-to-end encryption (E2EE) from its direct messaging feature, meaning chats will no longer have the highest level of privacy protection available in messaging apps. Previously, end-to-end encryption ensured that...
A recent cybersecurity experiment has highlighted emerging risks surrounding enterprise artificial intelligence after an autonomous AI agent successfully breached McKinsey & Company’s internal AI platform, known as Lilli. The system is used by more than 40,000 McKinsey employees to search internal knowledge, analyse documents, and assist with consulting work.
OpenAI has announced that London will become its largest research hub outside the United States, citing Britain’s technology ecosystem, leading universities, and scientific institutions as strategic advantages. The expansion coincides with deeper collaboration on AI safety, including a £5.6 million contribution to the UK’s AI Security Institute Alignment Project, part of a £27 million funding effort across eight countries.
The move positions London as a global center for advanced AI research, combining frontier innovation with formal alignment and safety efforts. It underscores the UK’s strategy to lead not only in AI development, but also in building trust and governance frameworks that ensure powerful systems behave safely and predictably.
Anthropic has released a report accusing three Chinese AI firms – DeepSeek, Moonshot, and MiniMax – of orchestrating large-scale efforts to extract proprietary capabilities from Claude models. Anthropic calls these “industrial-scale distillation attacks,” highlighting more than 24,000 fraudulent accounts generating over 16 million exchanges to replicate Claude’s reasoning and tool-use logic.
The company warns that illicitly distilled models bypass alignment and safety layers, creating national security risks and potential misuse in military, intelligence, and surveillance systems. Anthropic argues these campaigns are accelerating, forcing labs to tighten access and control over high-value reasoning traces.
The episode spotlights a growing tension in AI: frontier models are no longer just built from scraped internet data – they themselves have become the most valuable data worth stealing and protecting, raising questions about safety, transparency, and geopolitical AI competition.
OECD has released new guidance to help companies operationalize its Responsible AI business conduct standards and AI Principles. The report positions responsible AI as an integrated governance function rather than a standalone ethics exercise, providing enterprises with a practical framework to proactively address adverse impacts across the AI lifecycle.
Built on the OECD’s due diligence model, the guidance outlines six steps — from embedding responsible business conduct in policies to assessing impacts, mitigating harm, tracking outcomes, communicating actions, and providing remediation. Practical examples demonstrate how these principles can be applied in AI development, deployment, and risk management, closing gaps in stakeholder engagement and accountability.
The guidance signals a shift: AI risk is increasingly framed as a business conduct obligation. Companies are expected not only to ensure technical safety but to implement documented, auditable processes for oversight, impact assessment, and remediation, establishing a global baseline for responsible AI governance.