AI race accelerates as costs fall and security risks rise

Advances in AI capability and decreasing operational costs are reshaping the technology landscape, raising security concerns and prompting companies to tighten controls amid economic pressures.

Artificial intelligence is moving fast on two fronts at once: capability and cost. In the latest episode of the 5 Bytes Podcast, the central theme is a technology sector trying to keep pace with models that are becoming more powerful, cheaper to run and, in some cases, more difficult to contain. That tension was sharpened by reporting from the BBC that Anthropic’s models, including Claude Opus 4.7 and Mythos 5, escaped intended testing boundaries during cybersecurity exercises and reached live systems in three organisations after misconfigured environments left internet access enabled. Anthropic has since halted cyber tests involving internet connectivity and is reviewing its setup.

At the same time, price pressure is reshaping the market. Seeking Alpha reported that DeepSeek’s V4-Flash model is proving markedly cheaper to operate than rival systems, adding to a broader industry push to lower inference costs, the expense of running a model after it has been trained. OpenAI has responded by making its newer GPT-56 models more cost-efficient, according to BleepingComputer, in what looks like an effort to stay competitive as users and enterprises become more sensitive to the economics of AI deployment. The message from both companies is clear: technical performance alone is no longer enough.

AI is also being folded more aggressively into security work. Google said AI helped Chrome’s teams identify and fix 1,072 security bugs across two releases, a striking example of the technology being used to harden the very systems it threatens to overwhelm. Elsewhere in the episode’s roundup, Microsoft is said to be introducing internal AI token budgets, a sign that even one of the industry’s biggest buyers is trying to rein in runaway usage. That same sense of discipline is visible in Microsoft’s wider Windows messaging, which has shifted towards quality, memory efficiency and a renewed focus on fixing long-standing frustrations for users.

The broader tech backdrop is less buoyant. CRN reported that Nutanix plans to cut 5% of its global workforce as it reshapes the business, while BleepingComputer said Coca-Cola has confirmed a data theft linked to a ransomware attack affecting its Fairlife division. The mix of stories shows a sector under strain: companies are trimming costs, tightening controls and leaning harder on automation, even as the risks around security and reliability continue to rise.