AI Safety stories - Page 14

Anthropic launches Claude Sonnet 4.6 with major coding and computer-use upgrades, keeping pricing flat at USD $3 and USD $15 per million tokens.

Tailscale launches Aperture in open alpha, adding an identity-based gateway to log, govern and standardise workplace AI and agent use.

Mylobster.ai unveils plan for kernel-level AI assistants across major platforms, targeting a USD $200 billion-plus systems-level market.

Mainstream AI tools flunk detailed Australia and New Zealand K-10 curriculum tests, raising risks for teachers relying on chatbots in class.

Anthropic adds former Microsoft finance chief and US official Chris Liddell to its board as pressure mounts over AI governance and oversight.

Asia-Pacific organisations hit steeper technical hurdles than global peers as they try to scale agentic AI from pilot projects to production.

Anthropic raises USD $30bn in new backing, lifting the AI group's valuation to USD $380bn as demand for its Claude models surges.

OpenAI debuts GPT-5.3 Codex Spark, a Cerebras-powered research preview promising ultra-fast, real-time coding for ChatGPT Pro users.

OpenAI rolls out Lockdown Mode and new Elevated Risk labels in ChatGPT to curb prompt injection attacks and tighten data safeguards.

GitHub debuts Agentic Workflows, using AI agents with strict guardrails to automate repo chores while keeping maintainers in control.

Sue Ryder will deploy Heidi's NHS-approved AI scribe across UK hospices, aiming to cut paperwork and free clinicians for patient care.

AI-fuelled ransomware hit record levels in 2025, with BlackFog warning that around 86% of attacks worldwide are never publicly disclosed.

Proofpoint snaps up AI security specialist Acuvity to police 'agentic workspaces' as enterprises rush to deploy copilots and agents.

More consumers now treat generative AI as an everyday aide and even a companion, using it for routine decisions despite trust concerns.

Gartner warns a misconfigured AI in cyber-physical systems could knock out critical infrastructure and power in a G20 nation by 2028.

Wesfarmers is deepening its AI and cloud pact with Microsoft to roll out Copilot tools, agents and data platforms across its retail brands.

Daon warns viral ChatGPT caricature craze is handing fraudsters a richer dossier of personal data to fuel scams and social engineering.

Kyndryl launches policy-as-code guardrails to keep agentic AI workflows governed, auditable and compliant in tightly regulated sectors.

AI is turning romance scams into industrial-scale fraud, with deepfakes and chatbots driving billions in losses, Tenable warns.

AI-powered fraud rings industrialise online romance scams, fuelling USD $5.7 billion in losses and ushering in a 'dark age' for victims.