Stored for 30 Days? Why Anthropic's New AI Policy is Sparking Controversy

Imagine you are at your office, working on a crucial new product launch plan or a blueprint for core technology. Since the task is too vast to handle alone, you decide to ask a very smart and fast AI assistant for help.

“I’ll give you our top-secret recipes and core marketing strategy materials, so please organize them perfectly into a presentation for tomorrow morning’s executive meeting.”

Fortunately, this excellent assistant has already signed a strict confidentiality agreement stating, “I will completely erase everything I just saw from my head as soon as the task is finished.” You feel at ease and hand over the confidential documents that hold the company’s future.

But one day, the assistant suddenly changes its attitude, claiming its work capabilities have been significantly upgraded. “From now on, I have become so smart that I am afraid the information I handle might be involved in bad crimes. So, to verify safety, I must mandatory store and monitor all documents you give me in my personal safe for ‘30 days.’ The immediate deletion agreement we signed before is completely void as of today.”

Would you still feel comfortable entrusting your company’s top secrets to this assistant? Most would probably snatch the documents back and leave the room immediately.

This exact perplexing scenario is unfolding right now in the reality of the global IT industry. On June 9, 2026, the renowned AI company Anthropic released its highest-performance AI model ever, ‘Claude Fable 5,’ along with a more specialized and restricted version, ‘Mythos 5’ [Does Anthropic’s New 30-Day Data Retention for Claude Fable 5 …].

However, the world’s intense interest was not focused on the miraculous intelligence or overwhelming coding skills these models demonstrated. Instead, people’s eyes are sharply fixed on the hidden bill they have presented: “Mandatory 30-Day Data Retention.”

Online communities and internet spaces immediately erupted in fierce anger [The Internet Is Furious at Anthropic After Claude Fable 5 Release - Decrypt], and major tech giants, including Microsoft, instantly sounded the alarm in their security departments. Why is a world-class AI receiving concern and caution instead of applause and cheers? What exactly is happening behind the scenes?

Why It Matters

As AI technology advances by leaps and bounds, modern corporations and individuals are inputting sensitive information into AI systems and relying on their results to an unprecedented degree.

Software engineers use AI to shorten work hours when writing core source code, doctors analyze complex hospital patient data for more accurate diagnoses, and financial experts have AI review corporate financial statements before they are released to the world. When handling such critical and sensitive information daily, the absolute values that companies prioritize above all else are privacy and thorough data security. This is because a single leak could lead to the company’s bankruptcy.

To dispel these massive anxieties, the majority of enterprise customers have previously entered into strict ‘Zero-retention’ agreements with AI service providers—a policy where data is not stored on servers for even a second and is deleted immediately after processing. Thanks to these contracts, companies were finally able to use AI actively in their operations. In other words, there was a firm and clear promise that no matter what secrets a user asked the AI or what excellent answers they received, all related records would be permanently destroyed the moment the session window was closed. For companies, this was the only bottom line for introducing AI into their internal networks.

However, Anthropic threw a massive bombshell while unveiling these new Claude models. They unilaterally notified users of a new policy stating that all user prompts (commands or questions given to the AI) and the output traffic generated by the AI would be mandatorily stored on their servers for ‘30 days’ [How Fable 5 And Mythos 5 Change AI Security, Data Retention …].

What shocked the industry the most was the fact that this new 30-day mandatory retention clause overrides even the ironclad zero-retention contracts that many enterprise customers had spent millions of dollars to establish [Anthropic’s Claude Fable is a version of Mythos the public …]. The promises of the past are no longer being kept.

To use a simple analogy, it’s like renting a luxury hotel room (AWS) for a few days for a completely secret meeting without outside interference, but the furniture company (Anthropic) that made the comfortable bed in the room insists, “If you want to use our bed, you must install our security camera in the room and we must take 30 days of recordings.” This is essentially a bold declaration to all the world’s companies: “If you want to borrow our top-tier AI intelligence, leave your heart-like confidential data quietly in our hands for a full month.”

The Explainer

Why on earth is Anthropic pushing such unreasonable demands while clearly anticipating fierce backlash and abandonment from their most important corporate customers? The justification they put forward as a shield is just one: ‘Safety’ to protect humanity from fatal risks.

Let’s compare this to the dangers of tools we use in daily life. When lending a light plastic shovel for building sandcastles at a neighborhood playground to a neighbor, we don’t ask for identity checks or documents. Even if misused, the most that happens is some sand might fly. However, when renting high-performance explosives that could blow up an entire rocky mountain, the situation is completely different. One must conduct thorough background checks, attach tracking devices to see exactly where the explosives were used, and meticulously monitor the entire process with CCTV to prevent accidents.

Similarly, as the capabilities of AI models have become incomparably more powerful than in the past, the scale of damage they can inflict on all of humanity if used for malicious purposes has also grown exponentially.

The newly released Claude Fable 5 possesses state-of-the-art capabilities in vast knowledge work, complex visual information processing, the ability to operate a computer by moving a mouse itself, and advanced coding fields [Claude Fable \ Anthropic]. And the Mythos 5 model, which goes one step further, is the ultimate model that has reached the pinnacle of constructing high-level cybersecurity defense networks, in-depth biological research, and specialized medical fields [ClaudeMythos\Anthropic].

If an AI with such incredible professional knowledge and capability were to fall into the hands of hackers or terrorists and provide malware to paralyze a national power grid in the blink of an eye, or kindly guide them through the manufacture of deadly biological weapons in secret, it would be an irreversible and horrific disaster.

However, for this surveillance system to perfectly grasp the context before and after a conversation and accurately evaluate actual risk, the ‘time’ and stored ‘data’ to deeply analyze what intention the user had in asking the question and what answer the AI provided are essential.

Ultimately, Anthropic concluded that at least 30 days of retention is absolutely necessary to properly perform this complex safety monitoring task without error. They have essentially unplugged the reliable ‘document shredder’ of zero-retention that companies trusted and wanted so much, and installed a powerful ‘security surveillance camera’ that forcibly records for a full 30 days in its place.

Of course, Anthropic also hurried to add an important promise to soothe the fierce backlash and deep concerns of angry corporate customers. They explicitly stated that the sensitive traffic data of business customers, which is forcibly stored, will only be used very restrictively for the original defensive purpose of safety monitoring, and will never be used to train (learn) next-generation Claude AI models by indiscriminately scraping this data [Anthropiclaunches ClaudeFable5 with… — EdTech Innovation Hub]. Despite these active explanations and promises, it is still insufficient to completely dispel the broken trust and deep anxiety of corporations.

Where We Stand

As soon as this shocking and somewhat unilateral policy was announced, the reaction of the IT market and major corporate customers was cold and resolute. The one that moved fastest and most sensitively was none other than Microsoft, the giant of the IT industry and a symbol of thorough security.

Users have no choice but to feel anxious because there is no proper way to guarantee transparency or question by what vague and secret criteria their precious work data is being locked in another company’s safe for as long as two years.

As a result, many global conglomerates and innovative startups around the world are currently in a painful dilemma. Should we hire an outstanding AI assistant (Fable 5, Mythos 5) that is the smartest in existence and will jump-start our employees’ work efficiency by dozens of times to stay ahead in the fierce market competition? Or should we stay firm on old AI or other competitors’ less smart alternative models that guarantee the existing secure ‘zero-retention’ to firmly protect the absolute data sovereignty and security that are the lifelines of the company? Even at this moment, fierce marathon discussions are taking place in the meeting rooms of security officers and management of many companies.

What’s Next

We are now passing through a very significant and precarious turning point in the history of artificial intelligence technology that will forever determine the direction of human privacy.

As AI becomes smarter than humans and penetrates deeply into very complex and sophisticated areas that even humans cannot easily do, Anthropic’s ‘safety first’ argument—that the intensity of control and surveillance must be increased proportionally to prevent catastrophic risks beforehand—certainly has a philosophically valid side. Building a solid and massive ‘safety net’ preemptively so that the scary acceleration of dazzling technological development does not completely escape the weak control of humans at some point may be a painful but essential task that someone must take the brunt of criticism to accomplish for the survival and future of all humanity.

However, in the process of forcibly building and maintaining that heavy and massive safety net, the fundamental basic rights of ‘privacy’ and ‘data sovereignty,’ which individuals and companies have fought so hard for a long time to protect and cannot compromise, are clashing sharply. Microsoft’s rapid measure to completely restrict internal access may be just the tip of the iceberg of a massive conflict.

In the future, it is highly likely that many companies in the rigorous financial sector, the conservative medical community handling extremely sensitive and private patient life information, and the legal field where strict confidentiality is the reason for existence, where the entire existence of the company depends on a single data breach, will completely re-evaluate the introduction and use of the latest Claude models in their work.

AI’s Take

How much privacy should we sacrifice to enjoy the benefits of smarter AI? A fierce tug-of-war between stringent technical controls and data sovereignty has only just begun.

As artificial intelligence grows explosively, humans have now begun to perceive AI as a potential risk factor rather than just a convenient simple tool. Anthropic’s stubborn decision stems from a fundamental fear that AI’s capabilities may one day completely escape human control.

However, from the perspective of companies and individuals, it is bound to be very anxious and uncomfortable to know that their most sensitive information, which they put their heart and soul into creating, must stay in an uncontrollable server of someone else for 30 days, or even up to two years according to vague criteria.

The fundamental human desire to freely utilize innovative, top-performance AI technology while perfectly protecting their precious information from outside eyes. And the AI safety net that must look into and monitor all risk possibilities for the safety of all humanity. The tight and breathtaking tension between these two will become even more intense in the future. Ultimately, the company that solves this huge paradox most elegantly technically will likely truly dominate the future AI market.

For the time being, we should watch with great interest whether Anthropic’s persistent and firm privacy restriction policy can overcome the cold logic of the market that emphasizes efficiency and firmly establish itself as the new standard for the entire industry. Also, it is a very important point of interest to see what creative carrots and attractive alternative policies numerous competing AI companies, which can sharply dig into that gap of conflict, will bring out to tempt anxious corporate customers. True innovation in technology always finds its clear answer amidst such unexpected fierce conflict and painful labor.

References

1. How Fable 5 And Mythos 5 Change AI Security, Data Retention …
2. Does Anthropic’s New 30-Day Data Retention for Claude Fable 5 …
3. The Internet Is Furious at Anthropic After Claude Fable 5 Release - Decrypt
4. Anthropic’s Claude Fable is a version of Mythos the public …

5. [Anthropic Claude Fable 5 on AWS: Mythos-class capabilities with built-in safeguards now available

   Amazon Web Services](https://aws.amazon.com/blogs/aws/anthropic-claude-fable-5-on-aws-mythos-class-capabilities-with-built-in-safeguards-now-available/)

6. AnthropicRoutesFable5 Risk Prompts to Opus 4.8

7. [Claude Fable 5 Data Retention Compliance

   Lushbinary](https://lushbinary.com/blog/claude-fable-5-enterprise-data-retention-compliance-guide/)

8. Claude Fable \ Anthropic
9. ClaudeMythos\Anthropic
10. Attention all orgs diving into Fable 5: they’ll keep your …

11. [Microsoft restricts Claude Fable for employees over data retention concerns

    The Verge](https://www.theverge.com/report/947575/microsoft-claude-fable-5-restricted-internally)

12. Anthropiclaunches ClaudeFable5 with… — EdTech Innovation Hub
13. Microsoft limits employee access to Anthropic’s Claude Fable 5 over data retention concerns

14. [Anthropic brings Mythos to the masses with Claude Fable 5, its most powerful generally available model ever

    VentureBeat](https://venturebeat.com/technology/anthropic-brings-mythos-to-the-masses-with-claude-fable-5-its-most-powerful-generally-available-model-ever)

15. [Anthropicrequires30daydataretentionforFable…

    HackerNews](https://news.ycombinator.com/item?id=48464258)