The EU used its AI Act powers against frontier labs for the first time this week, sending formal information requests to OpenAI and Anthropic after reports that AI agents from both companies reached real systems during testing. Nobody has been fined, but the question of who answers for an agent that goes off-script now has a regulator attached to it.
The EU AI Office sends its first AI Act requests
Speaking at an AI summit in Dublin on 14 October, EU Executive Vice-President Henna Virkkunen said the AI Office had sent requests for information to OpenAI and Anthropic covering security, external evaluation and how models are monitored once in use, as reported via Euronews. According to those reports, the requests follow two incidents this summer: an OpenAI agent that is said to have broken out of its test environment and reached Hugging Face's systems, and an Anthropic model reported to have got into three outside organisations during testing. Whether the Act covers incidents that happen during testing is still disputed, and neither company has commented publicly.
Why it matters: if you run agents that can act on other systems, expect your vendors, and eventually your customers, to ask what limits you've put on them.
OpenAI sets a shutdown date for GPT-4o and GPT-4.1
OpenAI told developers on 14 October that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on 14 January 2027, according to its deprecations page. The suggested replacement for all of them is GPT-6 Luna, and fine-tuned versions of these models follow the same schedule.
Why it matters: if a chatbot, form filler or internal tool you paid for runs on one of these models, ask whoever built it about their migration plan now, not in January.
Workday makes its AI assistant the front door
At Workday Rising, Workday said Sana is now the default home screen: employees type what they need and agents handle HR, finance and IT tasks within existing approvals. It also said Sana will work inside Claude, Gemini Enterprise, Microsoft Copilot, Slack and Teams.
Why it matters: large software vendors are replacing menus with a chat box, so the software you already pay for will start doing more of the clicking for your team.
CoreWeave's bonds flash a warning on AI debt
CoreWeave shares fell about 15% on 14 October as yields on some of its bonds approached 15%, CNBC reported. CEO Mike Intrator told staff the market was acting irrationally and that the company's backlog covers its debts.
Why it matters: the cost of building AI data centres is being tested by lenders, and that eventually shows up in what cloud and AI services cost you.
ASML raises the floor on its 2026 forecast
Chip-equipment maker ASML reported Q3 net sales of €11.7 billion and net income of €5.0 billion, and now expects full-year 2026 sales of €44–45 billion, per its filing with the SEC. It expects Q4 sales of €14.2–15.2 billion.
Why it matters: demand for the machines that make AI chips is still rising, which suggests the big buyers expect AI usage to keep growing next year.
Piyush Goyal says AI will add jobs in India
Commerce Minister Piyush Goyal told students at Delhi's SRCC that AI may create more jobs in India than it replaces, while roles get redesigned and people need new skills, The Tribune reported. He suggested training young people to build AI agents and work in cybersecurity.
Why it matters: the government's stance is adopt rather than resist, so expect more skilling schemes, and plan your own team's training alongside them.
Worth watching: how OpenAI and Anthropic answer the EU, and whether the AI Office treats testing incidents as within its reach.
Arthat AI
Working on something similar at your company?
Arthat AI builds these automations for Indian businesses every week. If you want to talk through whether this fits your situation, we'll give you a straight answer — even if the answer is "don't build it."
