Between June and September 2026, all three of the large AI vendors released new top models, the protocol that connects AI agents to other systems was rewritten, and the EU's AI Act was amended. The National Library of Norway also released open Norwegian language models trained on licensed newspaper content. Below are the stories I think matter most for Norwegian businesses, with dates and sources. Statistics Norway (SSB) publishes new figures on AI use in Norwegian enterprises on 25 September, just after this was written. Last year, 30 percent of enterprises with at least ten employees used one or more AI technologies, up from 20 percent the year before.
The AI news that mattered this summer
Anthropic released Fable 5, then had to pull it for three weeks
What happened: On 9 June, Anthropic launched Claude Fable 5, its most capable model for general use. Three days later, the US government placed export controls on the model. Anthropic had no reliable way to check users' nationality in real time, so it suspended access for everyone. The model came back for users worldwide on 1 July. Its sibling model Mythos, which has fewer safeguards, remains limited to selected partners in cybersecurity and life sciences.
What it means for you: A service built on one model from one vendor can disappear overnight, for reasons neither you nor the vendor controls. If AI sits in critical workflows, you should know which model you fall back on, and you should have tested that it works.
Powerful models got cheaper
What happened: Anthropic launched Claude Sonnet 5 on 30 June and Claude Opus 5 on 24 July. According to Anthropic, Opus 5 comes close to Fable 5 at half the price. Fable 5.1 followed on 1 September with lower costs and fewer false refusals from its safety filters. On 22 September came Opus 5.5, which Anthropic says costs 40 percent less to run than Opus 5 on typical workloads.
What it means for you: The price of a given level of quality is falling month by month. An AI solution you rejected in the spring because it cost too much may be worth running the numbers on again. Choose the model based on the task rather than the release date.
OpenAI launched GPT-5.6 and GPT-6
What happened: On 9 July, OpenAI launched GPT-5.6 in three sizes (Sol, Terra and Luna), together with ChatGPT Work, a tool for businesses. In the first week of September came GPT-6 Astra, first to a limited set of organisations and then to paying ChatGPT users and through the API. OpenAI lists API pricing for Astra at 10 dollars per million input tokens and 50 dollars per million output tokens. The company added the smaller GPT-6 Sol and GPT-6 Luna on 22 September. Al Jazeera reported that the launch came amid growing criticism of safety around advanced AI agents.
What it means for you: If you use ChatGPT Business or Enterprise, your staff get new models without IT doing anything. Check which models are switched on, and whether your AI usage guidelines still fit.
Google updated Gemini Flash three times
What happened: Google released Gemini 3.6 Flash on 21 July, Gemini 3.7 Flash on 13 August and Gemini 3.8 Flash on 2 September. The long-awaited Gemini 3.5 Pro was still being tested with partners in July, according to TechCrunch, and I have not found a launch of it in the period.
What it means for you: Flash is Google's fast, low-cost model line, and it is what many companies use for customer service, search and automation. Three versions in six weeks means someone has to keep track of which version your solutions run on, and which older versions are being retired.
MCP got its biggest update yet
What happened: On 28 July, a new version of the Model Context Protocol was released. MCP is the standard AI agents use to connect to tools and data sources. The biggest change is that the protocol no longer needs a persistent session between agent and server. Authorisation has also been tightened. Anthropic, AWS and Microsoft described the release as an important step for running MCP at scale. Cloudflare wrote that MCP servers can now run on standard, scalable infrastructure.
What it means for you: MCP servers become simpler and cheaper to operate, and everyday tools such as firewalls and API gateways can read and route the traffic. If you have your own MCP integrations, some older features are now deprecated, with at least twelve months' notice before removal. Set aside time to upgrade.
The EU AI Omnibus came into force
What happened: The amendments to the EU AI Act entered into force on 27 July. Requirements for high-risk systems are postponed to 2 December 2027 and 2 August 2028. The transparency obligations in Article 50 still apply from 2 August 2026, but systems already on the market before that date have until 2 December 2026 to add machine-readable marking of AI-generated content.
What it means for you: If you offer AI solutions in the EU, the transparency rules apply already, such as the duty to tell users they are dealing with an AI system. We cover the regulation in more detail in a separate post.
Norway's AI law goes out for a new consultation this autumn
What happened: On 4 August, Norway's digitalisation minister Karianne Tung announced that the Norwegian AI law will go out for a new public consultation this autumn, with the EU changes built in. The aim is to present the law to the Storting in spring 2027.
What it means for you: Norwegian rules will follow the EU, but they do not apply yet. The consultation is a chance to speak up if something hits your sector badly.
The National Library released open Norwegian language models
What happened: On 10 August, the National Library of Norway launched the Borealis models in Bokmål and Nynorsk. They are trained on newspaper content licensed through an agreement with Kopinor, and the state pays NOK 45 million a year for the licence. The models are free to use for the public sector, research and business.
What it means for you: According to the National Library, this is the first time a country has made language models trained on licensed, copyrighted content freely available. That gives you Norwegian models where the rights to the training data are settled. I have not yet seen independent comparisons with the large commercial models, so test them on your own tasks. They may suit cases where you need to run the model yourself, for privacy or cost reasons.
What I would keep an eye on before Christmas
The vendors release new models faster than most businesses can evaluate them. Switching every time makes little sense. What matters is knowing what you use. Write down which models and versions sit behind each solution, and who is responsible for updating them. Make sure anything critical has a fallback with another vendor. Read the consultation on the AI law when it arrives, and look at the SSB figures on 25 September to compare yourself with other enterprises in your sector.
At Frontkom, I work on AI advisory and integrations, including MCP setups that connect AI assistants to the systems businesses already run.
Sources
- Redeploying Claude Fable 5, Anthropic, July 2026
- GPT-6 Astra: A new generation of intelligence, OpenAI, September 2026
- Release notes, Gemini API, Google, 2026
- The 2026-07-28 Specification, Model Context Protocol Blog, 28 July 2026
- AI Omnibus enters into force, European Commission, 27 July 2026
- Tung vil sende KI-loven med endringer på høring, Government of Norway, 4 August 2026
Transparency note: This article is based on the author's own analysis and experience. AI has been used as an editorial aid for language and structure.
