Verizon joins Project Glasswing to test Anthropic’s Claude Mythos model on its network

Verizon announced today that it is participating in Anthropic’s Project Glasswing, gaining early‑access to the Claude Mythos Preview model. The AI system is marketed as a specialist in finding complex vulnerabilities across operating systems, firmware, and critical software stacks, and Verizon plans to run it on its own infrastructure to harden the carrier’s network.

---

Why Mythos matters for a carrier

Claude Mythos is a large‑scale, safety‑focused language model that has been fine‑tuned on millions of security‑related code artifacts, bug reports, and vulnerability disclosures. Unlike general‑purpose LLMs, Mythos can:

• Parse binary blobs and firmware images to surface hidden backdoors that traditional static analysis tools miss.
• Correlate CVE data with proprietary codebases, suggesting mitigations that fit a carrier’s specific deployment patterns.
• Generate reproducible proof‑of‑concept exploits in a sandboxed environment, allowing security teams to verify a flaw before it is disclosed publicly.

Anthropic claims the model can reduce the time to identify a critical vulnerability from weeks to a matter of hours, a speed boost that is especially valuable for a network that supports billions of devices.

---

How Verizon will use Mythos

Verizon’s press release outlines a three‑phase rollout:

1. Baseline assessment – Mythos will scan Verizon’s internal code repositories, network‑function virtualization (NFV) scripts, and edge‑router firmware to produce a prioritized list of potential weaknesses.
2. Remediation testing – Security engineers will feed the model suggested patches; Mythos will evaluate their effectiveness and flag any unintended side effects.
3. Cross‑industry sharing – As part of Project Glasswing’s collaborative charter, Verizon will contribute anonymized findings to a shared knowledge base that other “global security leaders” can draw from.

All phases are governed by Anthropic’s safety protocols, which include strict data‑isolation, audit logs, and human‑in‑the‑loop review before any model‑generated code is deployed.

---

Ecosystem implications

Verizon’s involvement signals that telecom operators are beginning to treat AI‑driven vulnerability discovery as a core service rather than an experimental add‑on. The move could have several ripple effects:

• Competitive pressure – AT&T and T‑Mobile may accelerate their own AI‑security pilots to avoid falling behind in network protection capabilities.
• Vendor lock‑in considerations – By integrating Anthropic’s model into its security stack, Verizon will need to maintain a reliable API contract and possibly negotiate dedicated compute resources, which could tie the carrier to Anthropic’s cloud infrastructure for the foreseeable future.
• Regulatory visibility – U.S. telecom regulators have been watching AI‑enabled security tools closely. Verizon’s public safety commitments may set a benchmark for how carriers demonstrate responsible AI use.

---

What this means for consumers

For the average Verizon customer, the partnership is largely invisible, but the downstream effect is a more resilient network. Faster detection of zero‑day flaws means fewer service outages, reduced risk of data interception, and a stronger defense against large‑scale attacks that target the underlying transport layer.

---

Looking ahead

Project Glasswing is still in its preview stage, with only a handful of organizations granted access. Anthropic plans to open the program to a broader set of partners later in 2026, after the initial safety and performance benchmarks are met. If Verizon’s pilot proves successful, we could see AI‑powered vulnerability hunting become a standard component of carrier security operations.

---

For more details on Project Glasswing, see Anthropic’s official announcement and Verizon’s press release on the partnership.