Skip to content
Back to The Briefing

ChatGPT, Codex and OpenAI API: Invisible text watermarks arrive

By ··2 min read

Good morning, Tom, and welcome to Tuesday’s briefing. The Pentagon has stopped using Anthropic’s AI tools following its earlier blacklisting of the company, the BBC reports. OpenAI has also introduced invisible text watermarking: API customers can opt in now, with eligible ChatGPT and Codex output in the EU following over the coming weeks.

New releases

ChatGPT, Codex and OpenAI API. From 5 October, API customers worldwide can opt in to textGrain watermarks for selected models, off by default, while eligible ChatGPT and Codex text output across all plans in the EU will gain invisible watermarks over the coming weeks. More

Codex and ChatGPT Work. OpenAI’s Codex lead reports approximately 50% faster default generation for GPT-6 Astra and GPT-6.1 Sol through ChatGPT subscriptions, including Codex, Work and partner tools using Sign in with ChatGPT, with no settings change required. More

OpenAI API. From 5 October, authorised administrators of eligible API organisations with an established usage history can accept OpenAI’s standard Business Associate Agreement and enable HIPAA compliance support inside the API Platform, without an enterprise contract. More

Perplexity and Perplexity Computer. GPT-6.1 Sol is now available to eligible paid subscribers in Search and Computer, powering Light effort, while Claude Opus 5.5 powers Standard effort for eligible paid personal accounts. More

Perplexity Computer. Light, Standard, High and Ultra effort controls now extend to iOS 26.38.0 and Android 2.100.0 or later for Pro, Max, Enterprise Pro and Enterprise Max subscribers, following their earlier web release. More

Perplexity Computer. A newly documented DocSend connector lets users with Computer access and an eligible DocSend account analyse deal-room documents and viewing activity, subject to their organisation’s connection permissions. More

Training your team on ChatGPT? See our ChatGPT training.

Elsewhere in AI

The Pentagon has stopped using Anthropic’s AI tools after blacklisting the company as a supply chain risk in February over its refusal to remove safety guardrails, the BBC reports. Read

The Internet Watch Foundation says it has assessed 6,310 AI-generated child sexual abuse images in 2026, already 40% above last year’s total. Read

OpenAI apologised again for the Medicare hack at Australia’s parliamentary AI inquiry, saying it must do more to rebuild public trust. Read

Reflection AI claims its new Beam model matches a leading Chinese competitor, as competition over open AI models intensifies, the Financial Times reports. Read

Italian prime minister Giorgia Meloni is seeking to trademark her voice against AI deepfakes, becoming the first major politician to pursue such protection, the Financial Times reports. Read

Get The Briefing by email

Choose how often. We'll send a link to confirm.

How often