This week, two stories that affect you as a Dutch professional in different ways. Google is putting the latest Claude models behind a paid subscription in its programming environment. In the US, the government is opting for voluntary agreements with AI companies and a new task force, whereas in Europe legislation applies. What does that mean for you?
Claude Opus 5.5 and Sonnet 5.5 in Antigravity, but not for free
Google has added Claude Opus 5.5 and Claude Sonnet 5.5 to Antigravity, its AI environment for programming. Startup Fortune reports this, based on Google's own model documentation. An announcement in the changelog was still missing at the time. You only get access with a paid Google AI Pro subscription ($19.99 per month) or Ultra ($99.99 or $199.99 per month). With a trial subscription to Pro, the free version, AI Plus or Enterprise, you cannot use the new Claude models. According to the article, the older Claude 4.6 models will disappear from Antigravity on 2 November.
Are you using Antigravity with Claude 4.6 in your development team? Then check which subscription you have before 2 November. Without Pro or Ultra, you will only have the Gemini models left. If you work with clients' source code, also check under which terms your subscription falls. A consumer subscription is something different from a business environment with a data processing agreement. On /claude/ and /gemini/ you can see how both models score and what they cost. On /voor/softwareontwikkeling/ you will find the coding tools we have checked.
US: voluntary agreement and a Super Intelligence Force
At the end of September, President Trump and six leaders from the AI world signed the Joint Commitment on Frontier Responsibilities. They were Dario Amodei (Anthropic), Greg Brockman (OpenAI), Sundar Pichai (Google), Mark Zuckerberg (Meta), Elon Musk (xAI) and Jensen Huang (Nvidia), CDO Magazine writes. The companies promise "robust" controls on their models, co-operation with independent external auditors and regular consultation on shared safety standards. The agreement is not binding. It contains no sanctions, and it does not specify who will carry out the audits. Trump called it "morally binding". TechCrunch, by contrast, calls the agreement "deeply non-binding".
On 4 October, Trump also announced the Super Intelligence Force, SiliconANGLE reports. The group is led by intelligence chief Jay Clayton. It must report within 120 days on the risks and opportunities of AI and on the role of the federal government in that.
What this means for you
For your work in the Netherlands, these American agreements change nothing. The GDPR and the AI Act apply here. Since 2 August 2026, providers of generative AI must make their output recognisable. The rules for high-risk AI have been postponed until 2 December 2027, Recruitmenttech sets out. The promise of external audits sounds good, but without binding agreements you cannot rely on it for your own risk analysis or DPIA. So base your assessment on what providers commit to contractually: the data processing agreement, storage location and whether your data is used for training. For each model, this is listed on /veiligheid/. To see who supervises in the Netherlands, read /toezicht/.
In brief
- Antigravity with Claude? You need a paid Pro or Ultra subscription, and Claude 4.6 disappears on 2 November.
- The American safety agreement is voluntary and is no guarantee for your GDPR assessment.
- For you, the GDPR, the AI Act and the contract with your provider remain decisive.