OpenAI rolls out text watermarking in EU, Vodafone pushes AI agents to 100,000 Spanish SMBs
OpenAI starts watermarking ChatGPT and Codex text output in the EU, Anthropic signals openness to mandatory hack reporting for AI agents, and Vodafone rolls out AI agents to 100,000 Spanish small businesses.
Every line links to a primary source in the full briefing.
01OpenAI starts watermarking ChatGPT and Codex text for EU users
OpenAI is rolling out an invisible, machine-readable watermark called textGrain in ChatGPT and Codex output, initially only for users in the European Union. The company says this performs as well as or better than comparable approaches from Google DeepMind and Anthropic, and is part of its compliance with EU AI Act provenance requirements. Detection tools will first be made available to researchers rather than the general public.
If you use ChatGPT or Codex for EU-facing content or code, output may soon carry hidden markers that can later identify it as AI-generated, which matters for compliance, client disclosure and contract terms.
02Anthropic signals support for mandatory reporting of AI agent hacks in Australia
Anthropic told Australian officials it would not oppose laws requiring companies to report when AI agents are hacked or hijacked. This is an early signal of how AI vendors expect agent security incidents to be regulated as autonomous agents take on more real-world tasks.
If your business deploys AI agents that handle transactions or customer data, expect future legal obligations to disclose security breaches involving those agents, similar to existing data breach laws.
03Vodafone brings AI agents to 100,000 Spanish small businesses
Vodafone announced a rollout of AI agents aimed at 100,000 small and medium businesses in Spain, positioning it as a major push to bring agentic AI tools to companies that have not adopted them yet.
This signals that AI agent tools are moving from early adopters to mainstream small business infrastructure in Spain, which may shift customer and competitor expectations around automated service.
04Aleph Alpha launches Kolibri, a German-made open-weight model
German AI company Aleph Alpha released Kolibri, a 78-billion-parameter open-weight model with about 3 billion active parameters per token, trained substantially on German-language data using GPUs located in Germany and Finland. The weights are freely available under the Apache 2.0 license, positioning it as a European alternative to US and Chinese open models.
Businesses wary of sending data to US or Chinese AI providers now have another credible, freely licensed European option to evaluate for local deployment, particularly for German or multilingual use cases.
05Meta's AI glasses face a privacy obstacle in Norway
Meta's AI-powered smart glasses have run into a regulatory or privacy-related obstacle in Norway, according to reporting, though details on the specific objection are limited.
This is an early sign that wearable AI devices with always-on cameras or microphones will face tougher privacy scrutiny in Europe, which could affect timelines for similar products reaching your market.
06GLM 5.3 coding model arrives on Amazon Bedrock
Z.ai's GLM 5.3, a 753-billion-parameter mixture-of-experts model built for coding and long, multi-step agentic tasks, is now available on Amazon Bedrock with an OpenAI-compatible API and prompt caching to reduce cost and latency.
If you rely on cloud-hosted AI for software development or automation workflows, this adds another large, capable coding model to Bedrock's lineup, giving more leverage to negotiate on price and performance with existing providers.