Anthropic Just DESTROYED the Reason You Pay for Claude

Summary

Corey McClain argues that Claude subscriptions provide rental access to a powerful but provider-controlled tool, rather than ownership of software whose capabilities users can rely on indefinitely. His central concern is that Anthropic increasingly decides what kinds of work Claude will perform, and that restrictions may change with little warning.

Key points

Reverse Engineer Anything

The video describes Reverse Engineer Anything (REA) as an open-source project that runs on a user’s computer and connects Claude to software security research tools such as Ghidra and IDA. McClain cites rapid interest on GitHub and reports from users who said Claude stopped partway through reverse-engineering tasks.

Restrictions on game cloning

After AI-generated recreations of existing games drew complaints from publishers, Anthropic added instructions against creating copies of specific games—even under different names. The video says original games and genre-inspired projects should still be allowed. However, users reportedly encountered the restriction before a formal announcement and could not switch it off.

Broader safety filters

McClain presents the game-cloning restriction as part of a wider pattern. He cites safeguards that can block legitimate cybersecurity work, including reports of filters interrupting harmless tasks. He also notes Anthropic’s stated rationale: filters may miss genuine threats, so they intentionally allow a safety margin that can produce false positives.

Access controls

The video says access to more powerful cybersecurity capabilities has been divided into tiers. Some levels require an application, while the highest level is reserved for organizations.

Example involving an LG TV

McClain recounts a journalist’s report that Claude helped alter a rooted TV’s settings, then refused to help restore them using an exploit. He emphasizes that, in this account, the user wanted to recover their previous setup—not disable tracking.

Anthropic’s policy choices

The speaker discusses Anthropic’s positions on military work and its limits on mass surveillance and fully autonomous weapons. He frames these positions, along with product safeguards, as examples of a recurring question: Who should decide what AI tools may be used for?

The video acknowledges that Anthropic has also adjusted or expanded some policies and access, so it does not present the trend as entirely one-way.

What a subscription buys

McClain says that even high-priced plans come with usage limits and terms that allow features or access to change. His conclusion is that subscribers pay for access under policies set by the provider, rather than owning the underlying capability.

Proposed safeguard

McClain recommends keeping as much of a workflow as possible under the user’s control. Projects, tools, and processes should be preserved in portable forms so switching providers remains feasible.

Running local models can offer more control, though they may cost money and may not yet match the capabilities of closed models. McClain expects open and more efficient models to narrow the gap over time.

Personal example

McClain says Claude blocked him from documenting his own YouTube content workflow, interpreting his request as an attempt to extract reasoning traces.

Main speaker and sources

• Main speaker: Corey McClain.
• Sources discussed or quoted: Anthropic statements and policies; Anthropic employee Addi Osmani; REA developers and GitHub users; reports and posts about AI-generated game recreations; a Decrypt report about an LG TV; and government and court statements concerning Anthropic’s military contracts.

Practical takeaway

The video is primarily commentary and analysis, not a step-by-step guide or product review. Its main practical takeaway is to avoid depending entirely on any one AI provider and to keep workflows portable.