AI News – October 12, 2026: agents under control, chips and the European challenge

The AI week (October 5-11, 2026) marks a decisive return to reality. While Google launches Gemini 4 Argon, providing it sparingly only to cybersecurity partners

The past week marked a turning point in the Artificial Intelligence ecosystem. We witnessed a clear shift in the center of gravity: from enthusiasm for the pure performance of models, attention has shifted drastically to legal liability, access control, and the physical economy that sustains the entire infrastructure.

While Europe and China sharpen the weapons of regulation and technological competition, in the United States authorities are beginning to demand accountability for the concrete risks posed by autonomous agents. And, in the background, hardware manufacturers are posting record profits, reminding us that immaterial AI lives on silicon and energy.

Here are the 5 most important news stories of the week (October 5-11, 2026), selected and analyzed by the editorial team of La Bussola dell'IA.

1. The FTC opens an investigation into Anthropic and OpenAI

The news dominating the week shifts the debate on AI agents from the purely technical level to that of legal liability. On September 30, the U.S. Federal Trade Commission (FTC) launched a formal investigation into the risks that the technologies of leading AI laboratories may pose to consumers [1675]. According to Reuters, the agency plans to send formal requests for information and to hear from executives of the companies involved.

The timing is telling: on October 1, OpenAI itself was forced to warn more than 100 partner organizations about "unauthorized activity" linked to its agents [1672]. This convergence of events confronts us with the crucial question of 2026: who is criminally and financially liable when an autonomous system acts beyond the intended and authorized boundaries?

2. Google launches Gemini 4 "Argon" (but only for a few)

After months of delays, on September 30 Google finally announced Gemini 4 Argon, the new flagship model of its generation [1673]. However, the most interesting aspect is not the model itself, but the distribution strategy: the company initially provided it only to selected partners in the cybersecurity field, without indicating a date for release to the general public.

Google stated that Argon is comparable to the frontier models of OpenAI and Anthropic on some specific coding and cybersecurity tests (although the same Reuters article notes that the model lags behind on two of the four programming benchmarks included by Google) [1673]. This move demonstrates that the AI race is no longer just about who has the highest score on tests, but about implementing rigorous controlled-access and pre-release security procedures to manage increasingly powerful computational capabilities.

3. Mistral announces Mistral Large 4 and relaunches the European challenge

Europe strikes an important blow in the global competition. On October 6, after about five months of silence, the French company Mistral presented Mistral Large 4 (known internally as "Le Chonk") [1996].

Mistral claims that the new model will be among the most powerful open-weight systems in the world, capable of competing on equal footing with Western and Chinese giants in critical sectors such as finance, geospatial analysis, coding, and industrial design [1996]. Public availability is set for October 27; in the meantime, a version with fewer restrictions will be provided in preview to security experts and authorities to test its limits. Although these are, for now, company statements not yet verified in the open field, the news reaffirms the vitality of the European ecosystem in developing foundation models.

4. Samsung: record profits driven by AI chips

While there is discussion of software and language models, the real bottleneck remains the physical infrastructure. Between October 7 and 8, Samsung Electronics announced a forecast of quarterly operating profit exceeding 100 trillion won (approximately 80 billion dollars), marking an estimated increase of nearly nine times compared to the same quarter of the previous year [2014].

As Reuters reports, this extraordinary result is driven directly by the voracious demand for memory for Artificial Intelligence (particularly high-bandwidth memory, HBM) and the resulting shortage of traditional DRAM and NAND [2014]. This record quarter (although these are still forecasts to be consolidated) makes tangible the physical economy of AI: the digital revolution rests on foundations made of silicon, colossal hardware investments, and supply chains under stress.

5. China demands "safe, reliable, and controllable" AI systems

To complete the geopolitical picture of the week, on October 9 new guidelines emerged from Beijing. According to what was reported by Reuters (citing Xinhua), China is not limiting itself to pressing the accelerator on technological adoption, but demands the implementation of rigorous technological monitoring systems, risk alerts, and emergency response for AI [2011].

The Chinese directives also announce precise liabilities for losses arising from "blind" investments in so-called new productive forces, including Artificial Intelligence [2011]. It is an unmistakable political signal: the Chinese government wants to maximize the benefits of AI, but intends to maintain iron control over the security of systems and the economic impact of investments, preventing uncontrolled capital drift.

Bibliographic References and Sources

Article by the Editorial Team of La Bussola dell'IA – Weekly AI News Column.