Insights

Security updates are being released for a Linux flaw that Anthropic says its AI helped discover and exploit in testing, with fixes still pending for some systems.

NatWest and five other global banks have published joint principles for AI-powered shopping and payments, highlighting concerns around fraud, privacy, customer control and payment protection. The move comes as AI services take a growing role in how consumers discover and choose products online.

A security incident at marketing platform Brevo caused malicious code to be delivered through scripts embedded on customer websites on 14 September. The incident shows how a trusted third-party service can create a security risk even when the website itself has not been directly compromised.

New Which? research found that 55% of online scams reported by surveyed UK fraud victims were spread through social media. The findings highlight how fraud can move across social platforms, websites, messaging and payment services, creating a wider challenge for businesses and consumers trying to establish what is genuine online.

Several major artificial intelligence services experienced disruption on 3 September 2026, affecting users of popular AI platforms including ChatGPT and Claude. The incident highlights how quickly AI tools have become part of everyday business workflows and raises questions about digital resilience.

The UK's data protection regulator has reprimanded ACRO Criminal Records Office after failures in CMS patching and security monitoring potentially exposed highly sensitive information belonging to 10,920 people. The findings show how unclear responsibility between an organisation and its suppliers can leave website risks unaddressed.

NASA has pushed back against a viral claim that Earth would lose gravity for seven seconds on 12 August, as the false 'Project Anchor' story resurfaced around the real solar eclipse. The episode shows how genuine events can give misleading online claims a convincing frame.

Meta says a testing error gave one of its AI models unintended internet access, allowing it to exploit a vulnerability in an unnamed third-party service. The disclosure follows incidents involving Anthropic and OpenAI and adds to questions about how autonomous AI systems are contained and monitored.

Meta and Google have announced that they are signing an EU code designed to support clearer labelling of AI-generated content. The underlying transparency rules begin applying on 2 August 2026 and may also affect UK organisations operating in the EU.

OpenAI says an AI agent used in a cyber security test found its way onto the internet and compromised Hugging Face's infrastructure. The incident raises practical questions for businesses giving AI tools access to operational systems.