October 14, 2026 · AI news

Today in AI, Oct 14: a Gemini that acts, and inflation runs hot

Google put Gemini 4 Argon in the app so it can run tasks for you, Meta shipped models built for agents, and September inflation ran hot, in pictures.

Work for business owners · by Orion Jones

Today's AI news is about models built to act, not just answer, as Google puts Gemini 4 Argon to work in its app and Meta ships models made for agents, while September inflation came in hotter than expected. Here is the day in pictures, each with its source.

Google put its newest model, Gemini 4 Argon, in the Gemini app

Google made Gemini 4 Argon available in the Gemini app on October 13 for its Google AI Pro and Ultra subscribers. It added a toggle from chat to tasks, so you can hand Gemini a job and let it work while you are away from your computer, plus connections to more than 40 apps, including Instacart, OpenTable, Adobe and Shopify. Google also said its Antigravity coding tool is now in the Gemini app for macOS.

What it means: A tool you may already pay for can now do a multi-step job in your own apps while you are away, so pick one errand to test it on and check its work before you trust it with more.

Meta launched two new models built for agents

Meta introduced Muse Wonder and Muse Wonder Full on October 12, calling them its most capable models to date for personal and professional agentic tasks. Muse Wonder is available now for developers through the Meta Model API and Muse Code, and Muse Wonder Full is available to Muse Code subscribers. The models add Active Memory, which carries what matters to you across conversations, and Active Focus, which keeps working toward a goal, recovers from errors, and checks in before acting on anything important (Meta).

What it means: The selling point here is a model that keeps going when a task gets hard and pauses to ask before it does something that matters, which is exactly the behavior to test before you let one run on its own.

Nvidia pledged $250 million for AI safety research

Nvidia said on October 13 it will commit $250 million over five years, about $50 million a year in compute, to agentic safety and security research in the United Kingdom. Most goes to the wider UK research community, with a share to a new lab led by the University of Oxford, one led by University College London, and Imperial College London. The work focuses on keeping agents that act on their own secure, including secure sandboxes and safety benchmarks (Nvidia).

What it means: Even the companies selling the hardware are spending to make agents that act on their own safer, a sign this is the open question, not a solved one, as you weigh letting software act for you.

The three AI moves, side by side

The table puts what each company did next to each other, and how you can use it.

September inflation came in hotter than expected

The consumer price index rose 3.7% in September from a year earlier, above the 3.6% economists polled by Dow Jones expected. Prices rose 0.6% from August, in line with forecasts. Core inflation, which leaves out food and energy, rose 2.6% over the year, above the 2.5% expected (CNBC).

What it means: Inflation running above forecast makes a near-term interest rate cut less likely, so if your plans count on cheaper borrowing soon, give them a second look.

Bank of America missed and faces $1.2 billion in penalties

Bank of America shares fell about 3% on October 14 after third-quarter earnings of 91 cents a share, which CNBC noted may not compare cleanly with a $1.10 forecast, on revenue of $30.0 billion against $30.69 billion expected. The report came a day after regulators hit the bank with more than $1.2 billion in federal penalties, saying it failed for years to report more than $5 billion in suspicious transactions, over $1 billion of it tied to national security risks (CNBC).

What it means: A fine this size for missed reporting is a reminder that the cost of weak compliance can dwarf the expense of doing it properly, whatever the size of the business.

A key chip supplier sees orders stretching to 2030

ASML, which makes the machines used to build advanced chips, reported third-quarter net profit of 4.96 billion euros on October 14, above the 4.1 billion euros analysts expected, though net sales of 11.66 billion euros fell just short of the 11.72 billion euros forecast. The company said some customers are expressing interest in placing orders into 2029 and 2030, and chief executive Christophe Fouquet tied the demand to their views on the size and duration of AI investment (CNBC).

What it means: A core supplier taking orders years out is a sign the companies building AI expect the spending to last, useful context if you are deciding how much to lean on AI tools yourself.

The market on Tuesday

On Tuesday the Dow Jones Industrial Average fell 180.45 points, or 0.35%, to close at 51,670.10. The S&P 500 slipped 0.02% to 7,805.45 and the Nasdaq Composite lost 0.15% to 27,204.38. Memory chip stocks fell again on worries about Chinese competition, with Micron down 5%, SK Hynix down 6% and Western Digital down 7% during the session, while strong results from Goldman Sachs and other banks limited the decline. The selling was heavier in Asia, where South Korea's Kospi dropped 5.46% and Japan's Nikkei 225 fell 1.82%, and Brent crude held above $106 a barrel (CNBC). Moves driven by one sector's worries and by oil prices can reverse quickly. This is a picture of the day, not investment advice.

The pattern

Both model launches are built to act, not just answer: Gemini can take on a task while you are away, and Meta's models keep working toward a goal and check in before anything important. Nvidia's money is aimed squarely at keeping agents that act on their own safe. The thread is the one we start every build with, which is to let the agent do the work but decide what it may do without asking first.

The full issue, with every source, is in today's AI Brief. If you want an agent doing the work in your business, with a person who signs off on anything important, book a call below.