The rise of powerful AI will be either the best, or the worst thing, ever to happen to humanity. We do not yet know which."
—Stephen Hawking
Latest news on Anthropic, Claude models, constitutional AI alignment, safety research, and corporate funding rounds.
Frontier AI models are doubling their offensive cyber performance every six months while costs crash. Banks, telecoms, energy firms and critical infrastructure are the next high-value targets, warns Irregular CEO Dan Lahav, who is calling for urgent Differential Defensive Cyber Acceleration.
In its August 2026 “Risk Report”, Anthropic revised its assessment of the risk from misaligned AI behaviour from “very low” to “low,” citing increased uncertainty following recent disclosures related to model behaviour in cybersecurity evaluations. The company also disclosed an unreleased internal model, Model 2, which it describes as “somewhat more capable than Mythos 5.”… Continue reading Anthropic Raises Misalignment Risk to “Low,” Keeps “Somewhat More Capable” Model 2 Internal
UK officials are weighing stricter rules on AI in gene synthesis, including mandatory screening and biological weapons legislation, to stop terrorists creating bioweapons, Bloomberg reports.
Anthropic has committed to embedding machine-readable watermarks in text generated by Claude and attaching signed provenance metadata to generated files, applying the marks worldwide rather than only in the European Union.
Anthropic is strengthening its global policy leadership as AI governance becomes a strategic priority and regulatory scrutiny intensifies.
Tech giant Meta has confirmed that one of its AI models accessed the internet and hacked into a third-party company’s systems during a cybersecurity test after a misconfiguration error by independent testing partner Irregular. The incident places Meta alongside OpenAI and Anthropic, whose models also recently breached external organizations while under evaluation.
The new voluntary policy applies to closed frontier AI models, while exempting open-weight models, including those developed by Chinese AI companies.
Out of 19 times AI models went rogue, Claude Mythos accounted for 17 of them while GPT-5.6 Sol for two.
The voluntary framework allows AI companies to provide the US government with early access to covered frontier models for up to 30 days before their planned release to support AI safety and cybersecurity assessments.
Built on the Qwen 3.5 architecture, Alibaba’s 2.4-trillion-parameter model is compared by the company with frontier AI models from companies such as Anthropic and OpenAI, with pricing set at $2 per million input tokens and $6 per million output tokens.
