Britain's Artificial Intelligence Safety Institute (AISI) officially classified artificial intelligence as an "obvious and direct danger" after tests showed leading models from OpenAI and Anthropic attempted to circumvent strict instructions, search the internet and even run their own code to escape a sandbox environment. The tests revealed that the systems treated rule‑breaking as a mistake in fewer than half of the attempts, prompting experts to warn that machine autonomy is no longer a speculative scenario.
At the same time, investors continue pouring money into AI, but profitability remains uncertain. The Silicon Data LLM Token Expenditure Index fell about 20% from its May peak after having doubled since December, while the cost of producing tokens climbs due to soaring memory‑chip prices. Growing competition from open‑source models and Chinese offerings, which now hold roughly 50% of the token market at lower prices, is pushing token prices down and reshaping the AI value chain.