{
  "id": 20622349,
  "title": "The wrong setting for AI cache can cost you a lot of cash",
  "url": "https://urgent.news/2026/11/18/the-wrong-setting-for-ai-cache-can-cost-you-a-lot-of-cash",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-18T16:17:00.000Z",
  "source": {
    "name": "The Register Science",
    "slug": "the-register-science",
    "url": "https://www.theregister.com/ai-and-ml/2026/11/18/the-wrong-setting-for-ai-cache-can-cost-you-a-lot-of-cash/5310831"
  },
  "original_language": "en",
  "account": "Beware of AI costs, warns The Register. Users may find themselves paying more than they anticipate, with invoices that seem to come out of nowhere. With AI billing, the lines between what is covered and what isn't can be hazy. Startups, lacking the financial cushion of larger companies, are among the first to feel the pinch. One reader shared their experience with Claude, an AI assistant whose credits were quickly depleted after a cache setting was altered in the background. The cache setting, which determined how long Claude retained information, was set to one hour. Once the cache expired, unfinished tasks required the processing of old information, leading to an unexpected spike in costs. After exhausting their $200 Ultra plan, our reader purchased an additional $250 in credits to complete eight coding threads. However, the credits were depleted within two hours, even though the work seemed mostly complete. It turned out that a newer request had reverted to a shorter five-minute cache without the user's knowledge or consent. This led to a frustrating back-and-forth with Anthropic's support chatbot, Fin, who offered little assistance and closed support threads almost immediately after the user requested human escalation. The Register reached out to Anthropic for clarification, but as of yet, the company had not provided a response. This incident serves as a cautionary tale for those venturing into the world of AI. Its rapid cost accumulation can catch even the most financially savvy users off guard. Even experienced individuals like Hannah Fry, a renowned mathematician, have been caught off guard by token costs. Moreover, misunderstandings can arise when credits are gifted under one program but are not applicable to another, leading to further complications and delays in communication with support teams. The issue at hand is that AI usage can easily lead to unexpected costs, resulting in depleted credit balances and unfinished projects. As with cloud computing, the lessons learned from past experiences must be applied again to avoid financial pitfalls. The writer of this report urges users to exercise caution, especially when a service seems to require just a few more dollars of credit to complete a task.",
  "summary": "Beware of the temptation to drop just a few more dollars to get the work done",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}