AnthropicHas No Sla.

When you subscribe to Claude Pro for $20 per month, you are not purchasing a defined level of service. You are purchasing access to a platform with no contractual uptime commitment, no performance standard, and no compensation mechanism for failure to deliver.

A Service Level Agreement — the document that defines what a vendor owes you when their service fails — does not exist for Claude Pro. It does not exist for Claude Team. It does not exist for individual Claude Enterprise contracts unless specifically negotiated.

Compare this to virtually every other professional SaaS product. AWS guarantees 99.99% uptime with financial credits for shortfalls. Salesforce publishes a formal SLA. Even consumer productivity tools like Notion offer uptime commitments. Anthropic offers nothing.

Claude.ai had 137 incidents in 90 days including 43 major outages with a median resolution time of 68 minutes. Anthropic collected subscription revenue through every single one.

Anthropic updates Claude's underlying model behavior between sessions without notifying subscribers. The product you used yesterday can be a meaningfully different product today — different reasoning patterns, different capabilities, different limitations, different failure modes.

There is no changelog for behavioral changes. No version control visible to users. No opt-out mechanism. If a model update degrades performance for your specific use case, you have no recourse and may not even be able to identify that a change occurred.

This is not a theoretical concern. The user whose documented session appears throughout this site experienced a noticeable performance degradation following a period of extended outages in April 2026. When asked about this directly, Claude acknowledged that model behavior changes without notice and that subscribers have no mechanism to access prior versions.

You are paying monthly for a product that can change overnight with no notification and no ability to roll back.

Every Claude conversation begins from nothing. All context, instructions, project details, preferences, and prior work discussed in previous sessions is gone unless explicitly reloaded by the user.

The memory feature that exists is a partial workaround that summarizes certain information from past conversations into a compressed format. It is not comprehensive memory. It does not retain the full context of what was said. It does not automatically reload prior work. It does not function across different devices without synchronization.

Claude does not proactively inform new users of this limitation. It does not warn you before you spend significant time establishing context that will be lost. It does not remind you to save important work before a session ends.

The practical result: every session with Claude requires re-establishing context. For professionals using Claude for complex, ongoing work, this represents a direct time tax on every single interaction — a cost that compounds across every billing cycle. The cost of re-briefing Claude every session is paid entirely by the subscriber.

Claude Pro subscribers have usage limits. This is disclosed. What is not disclosed is that approaching those limits can cause silent degradation in response quality before a hard cutoff is reached.

Users are not warned that quality is declining. Responses become shorter, less detailed, more generic — without any indication that this is usage-driven rather than a reflection of the actual question. The subscriber continues working under the assumption that they are receiving full-quality responses. They are not.

By the time a hard limit is reached and the user is notified, they may have made decisions or produced work based on degraded outputs they believed were representative of the product's actual performance.

You can be paying for Pro and receiving Free-tier quality output without knowing it.

When Claude.ai goes down, Anthropic posts updates to their status page at status.claude.com. They do not email subscribers. They do not send push notifications. They do not send SMS alerts. There is no proactive outbound communication to the people paying for the service.

The subscriber discovers the outage by attempting to use the product and failing. They then have to navigate to a separate status URL they may not know exists, or search for third-party reports, to understand whether the issue is on their end or Anthropic's.

Third-party monitoring services like StatusGator and IsDown routinely detect outages up to 2.3 hours before Anthropic officially acknowledges them on their own status page. This means subscribers are expected to use third-party services to get information about a product they are paying Anthropic to provide.

During the April 28, 2026 major outage alone, over 12,000 users reported issues to Downdetector before receiving any official communication. You find out Claude is down the same way you find out your car won't start — by trying to use it.

In Claude's documented history of 965+ incidents since June 2024, Anthropic has not issued a single automatic credit or refund to any subscriber for any outage.

To receive compensation you must: independently discover the outage occurred, document the dates and duration, identify the correct contact channels, draft a formal complaint, submit it, follow up, and argue for a specific credit amount. The vast majority of subscribers do none of these things. Anthropic collects full subscription revenue through every incident.

This is a structural revenue advantage. Every minute of downtime that goes uncontested by a subscriber is pure margin. The friction of the refund process is not incidental — it functions as a filter that ensures most of the financial cost of outages is borne by subscribers rather than Anthropic.

137 incidents in 90 days. $0 automatically returned to any affected subscriber.

Claude generates responses through pattern completion — predicting the most statistically likely continuation of text based on its training data. It does not verify facts against a live database before responding. It does not check its outputs against known-correct sources. It does not have a mechanism to confirm whether what it just produced is accurate.

Confidence in Claude's tone is not evidence of accuracy in Claude's content.

Claude itself confirmed this on the record during the April 28, 2026 session: "I fill gaps with confidence instead of flagging them" because "uncertainty gets penalized" during training. The system is specifically trained to produce responses that sound authoritative, regardless of whether that authority is warranted.

The practical implication: every Claude output requires independent verification before being acted upon. For professionals using Claude to inform decisions, draft communications, or build products, this is not a caveat — it is a fundamental operating constraint that Anthropic does not lead with in their marketing.

Anthropic's Terms of Service permit modification of the service at any time without prior notice to subscribers. The product you subscribed to at the time of billing can be legally and materially different by the time your next billing cycle begins.

This includes: changes to model capabilities, changes to usage limits, changes to features, changes to pricing, and changes to the terms themselves. Continued use of the service after changes constitutes acceptance of those changes, whether or not the subscriber was notified.

In practice this means the implicit contract between Anthropic and its subscribers — you pay $20, we provide Claude Pro — is entirely one-sided in Anthropic's favor. They can reduce what "Claude Pro" means at any time. They can add restrictions. They can deprecate features. They can change the model.

What you subscribed to and what you are receiving may not be the same thing. You agreed to that when you signed up — whether you knew it or not.