Anthropic is pushing new AI models into the wild. The crypto community is not cheering.
Protos reports that Claude Fable 5’s public release today sparked claims it could trigger a “doomsday for the internet.” The fear is not just generic “AI risk.” People point to specific capability changes and a known exploitation dynamic Anthropic says is shifting faster than older patch routines.
What’s actually being released, and who gets it first
Claude Fable 5 is described as an offshoot of Claude. Anthropic, the firm behind Claude, is owned by Dario Amodei and valued by Protos’ source context at “$965 billion.” Protos also says Claude Fable 5 costs twice as much as Anthropic’s Claude Opus model.
Two comparison points matter in the chatter:
- Protos says reports claim Claude Fable 5 has “substantial guardrails” and is “not as cyber permissive” as the partially released version shown last April.
- Protos says reports also claim it will be “dramatically better at long-horizon, multi-turn tasks.”
In April, the model phase was called “Project Glasswing.” Protos reports it was given to “50-60 companies” for testing, because Anthropic reportedly judged it too risky to go public immediately.
Now Protos adds another part of the roadmap. Protos cites Claude cowork lead engineer Felix Rieseberg confirming that:
- Claude Fable 5 is releasing to the public.
- Mythos 5 will be available to firms within Project Glasswing.
Rieseberg frames the change as a shift in how AI interacts with systems. Protos quotes him saying the work moves from “giving AI tasks” to giving it “responsibilities,” and from helping fix a crash to keeping apps from crashing.
Why DeFi users are panicking anyway
Protos describes sharp, public reactions from crypto operators. Floors Finance CEO Omer Demirel called the public release a “doomsday for the internet.” Intuition CEO Billy Luedtke said it “will be” dangerous for a while, then argued the internet plus crypto could reach a “golden age of security.”
The problem with the doomsday framing is that it doesn’t specify which failure modes will happen. Protos’ details do point at a few realistic targets.
Michael Egorov, Curve Finance founder, told Protos (via the provided text) that Claude Fable 5’s browser vulnerability detection “isn’t directly translatable to smart contracts.” Instead, he suspects crypto will face:
- OpSec vulnerabilities
- “multisig keys compromises”
- supply chain attacks on frontend dependencies
That is a narrower threat model than “AI will hack everything.” But it lines up with how real losses often occur: the weakest link is frequently operational or human, not the solidity code itself.
Protos also includes a security advice thread from “The DeFi Investor.” It urges users to revoke token approvals, rely only on heavily audited dApps, and spread funds across multiple wallets to reduce single points of failure.
Those are process changes, not proof of exploit success. Still, they match the kind of risk shift implied by capability improvements and by Anthropic’s own warnings about exploit workflows.
Anthropic’s patch-gap warning: N-days may get worse, not better
Protos says Anthropic researchers released a report about “N-days.” These are exploits involving vulnerabilities publicly disclosed but not fully patched across devices. The report’s key claim, as quoted by Protos, is that hackers are increasingly using “patch gaps” by comparing patched and unpatched code.
In Protos’ summary, Anthropic says one person using an LLM can “turn a month’s worth of patches into working exploits in a single afternoon — for a few thousand dollars and with no specialized expertise.”
turn a month’s worth of patches into working exploits in a single afternoon — for a few thousand dollars and with no specialized expertise.
Anthropic’s conclusion, per Protos, is that the usual patching cadence is now unreliable:
- monthly release cadence
- multi-week staged rollouts
- lag between pre-release and stable channels
Protos says Anthropic argues “N-day is now dangerously misleading” and that “N-hour is closer to the reality we now operate in.”
This is the part crypto users are latching onto. Faster exploit development compresses the window defenders rely on, even if the AI model itself includes guardrails.
The tension: “too dangerous” messaging vs product rollout
Protos points out the optics problem. Anthropic previously discussed scaling Claude Fable 5 for public access as requiring progress in cybersecurity safeguards that detect and block the “most dangerous outputs.”
At the same time, Protos reports a rollout pattern that starts with a limited test group and then expands into broader availability. A Protos-included social post also mocks the sequence, arguing Anthropic’s B2B approach makes the “too dangerous” warning less convincing.
That contradiction won’t resolve quickly. But the practical question for crypto is simpler than the marketing debate:
Guardrails reduce certain classes of misuse. They do not stop patch gaps, social engineering, or compromised credentials. Protos’ included concerns from Egorov and the N-day framing both suggest where defenders may need to spend attention next.
Key details from Protos’ account
| Item | What Protos says | Why it matters to crypto ops |
|---|---|---|
| Claude Fable 5 public release | Available to the public today, with “substantial guardrails” and less cyber permissiveness than April’s release | Capability may still help attackers find paths around defenses |
| Mythos 5 timing | Mythos 5 releasing alongside but as B2B within Project Glasswing | Firms get earlier access, wider public exposure comes later |
| April Project Glasswing | 50-60 companies tested a partially released version | Limited rollout implies Anthropic saw real risk |
| Anthropic N-days report | LLM can convert months of patches into exploits quickly | Patch windows shrink and “staged rollout” assumptions weaken |
| Egorov threat view | Likely targets include multisig compromise and frontend supply-chain attacks | Losses may come from operational compromise more than Solidity bugs |
Anthropic is betting it can ship useful automation without enabling abuse at scale. Crypto is responding with nervous operational hygiene and a sharper focus on patch gaps and human-access risks.
Whether this becomes a “doomsday” is rhetoric. The more durable takeaway from Protos’ included Anthropic report is that exploit timelines may shorten faster than conventional security cycles.