AI Safety stories

Xero unveils AI-native finance platform and JAX assistant to cut admin for small firms, with accountants and bookkeepers also in line for gains.

Anthropic's Mythos spots corporate network attacks in hours, while security experts warn unmanaged AI agents are becoming a critical enterprise risk.

Areto founder Jacqueline Comer wins inaugural global AI safety prize as the New Zealand moderation firm expands its fight against online harm.

OpenAI expands its enterprise AI push with a biology-focused model aimed at speeding target discovery, analysis and early drug development.

OpenAI expands Trusted Access for Cyber with Bank of America, BlackRock and others, backing defenders, researchers and open-source security teams.

Anthropic rolls out Claude Opus 4.7 with sharper coding, stronger image handling and new cyber safeguards, plus fresh API controls and review tools.

Testlio rolls out human-led AI chatbot testing as brands face pressure to prove safety, accuracy and trust before customer use.

Mastercard and Lobster.cash to let OpenClaw users authorise AI agents to spend on existing cards, backed by Verifiable Intent.

Survey finds AI agents are spreading faster than security teams can track them, with most firms lacking visibility, rollback controls and identity governance.

Human-led AI security gains ground as experts warn autonomous SOCs risk hidden errors, weaker accountability and costly business disruption.

Emergent unveils Wingman, an always-on AI agent for email, scheduling and sales support as Australian firms seek to ease admin pressure.

Emergent's Wingman puts autonomous AI agents to work across email, calendars and chats, with human approval for higher-risk actions.

AI testing lags as more than half of organisations ship features, with 52% saying fewer than half of projects reach full production.

Survey finds 77% of security professionals want AI tools in security operations kept under human oversight, as 88% adopt guardrails.

Forrester says the next wave of AI will reach robotics, transport and commerce, with agentic tools, security and trust set to pay off first.

TrendAI and Anthropic join forces to embed Claude in Vision One, targeting AI vulnerability research and automated cyber defences.

Thoughtworks warns AI-assisted coding is swelling software complexity, as developers lean on older controls to curb security and oversight risks.

OpenAI broadens Trusted Access for Cyber to verified defenders, giving vetted users GPT-5.4-Cyber for tougher security work and code analysis.

Anthropic strengthens its board with Novartis Chief Executive Officer Vas Narasimhan, giving its trust-appointed directors a majority.

Denodo survey finds real-time data, security and context are the main barriers as firms struggle to trust agentic AI at scale.