1. Overview of the incident: U.S. executives urgently summoned banking giants to deal with new AI risks

In early April 2026, U.S. Treasury Secretary Scott Bessent and Federal Reserve Chairman Jerome Powell urgently convened the CEOs of a number of systemically important banks at the Treasury headquarters in Washington, D.C., to discuss the cybersecurity risks posed by Anthropic's latest AI model Mythos (Claude Mythos Preview). The closed-door meeting was hastily arranged around April 7, with participants including Citigroup CEO Jane Fraser, Morgan Stanley CEO Ted Pick, Bank of America CEO Brian Moynihan, Wells Fargo CEO Charlie Scharf and Goldman Sachs CEO David Solomon. JP Morgan CEO Jamie Dimon was unable to attend for some reason.

The background of the meeting is that Anthropic announced the Mythos model around April 7th. This model has demonstrated strong network capabilities in internal tests. It can not only discover thousands of high-severity vulnerabilities, but also independently exploit these vulnerabilities and chain multiple weaknesses into complex attacks. Anthropic made it clear that the model's capabilities have exceeded that of most human security researchers, so it decided not to release it publicly, but to make it available to about 40 selected partners on a limited basis through the "Project Glasswing" project for defensive vulnerability repair rather than offensive applications.

This move quickly attracted the attention of high-level officials in the U.S. government. The Treasury Department and the Federal Reserve want to ensure banks are aware of potential risks and take necessary protective measures. Powell has previously publicly mentioned the threat of cyberattacks to the financial system, and Bessent's participation highlights the Trump administration's emphasis on the risks of this emerging technology. The content of the meeting has not been made public, but multiple media reports indicate that the focus is on preventing the damage that AI-assisted hackers may cause to critical infrastructure and financial networks.

2. Technical capabilities and dual-use risks of the Mythos model

Anthropic disclosed the performance of the model in detail in the system card and blog accompanying the release of Mythos. It has achieved a significant leap forward in software engineering and network security tasks, and can autonomously discover and exploit zero-day vulnerabilities (zero-day), including weaknesses that have long gone unnoticed by humans. For example, the model found a 27-year-old vulnerability in OpenBSD that can remotely crash the computer through a simple connection; it also discovered multiple vulnerabilities in the Linux kernel that can upgrade ordinary user rights to full control.

Testing showed that Mythos identified a large number of high-severity vulnerabilities in thousands of open source software stacks that had passed millions of automated reviews without being discovered. The model can not only locate problems, but also write complex exploit code, escape the sandbox, hide traces, and even implement advanced techniques such as JIT heap injection in the browser sandbox. Anthropic admitted that these capabilities have a dual purpose: on the one hand, they can be used for defense to speed up the patching of critical software; on the other hand, if they fall into the hands of malicious actors, they may accelerate attacks on critical infrastructure such as hospitals, power grids, and power plants, or target the penetration of financial systems.

Experts have divided opinions. On the one hand, AI security researcher Roman Yampolskiy and others warned that such models will accelerate the development of new threats such as hacking tools and biological weapons, making it difficult for humans to keep up. On the other hand, some analysts believe that Anthropic may have a marketing component and its actual capabilities need to be independently verified. However, the autonomy shown by the model in controlled testing is alarming enough. Anthropic also hired clinical psychologists to evaluate the model's "personality," and the results showed that it has high impulse control and reality testing capabilities, but the company still emphasized the uncertainty of the subjective experience of AI.

Anthropic founder Dario Amodei previously wrote that humans are about to gain "almost unimaginable power", but there are still questions about whether social, political and technological systems are mature enough to control it. This is highly consistent with the discussion triggered by Mythos: the core risk is not a sci-fi AI rebellion, but the misuse or loss of control of powerful tools.

3. Entanglements with the U.S. Government: Supply Chain Risk Designation and Legal Disputes

The Mythos incident occurred against the backdrop of Anthropic's tense relationship with the Trump administration. Earlier in 2026, the Pentagon designated Anthropic a "supply chain risk" due to its refusal to remove the model's security limitations around autonomous weapons and domestic surveillance. The designation typically targets foreign adversaries and is intended to limit their work with defense contractors. Anthropic subsequently sued, alleging retaliation, and won a partial preliminary injunction, but a federal appeals court recently denied its request to halt the designation.

Despite the legal dispute, Anthropic said it has discussed Mythos' offensive and defensive cyber capabilities with U.S. officials. The Pentagon has previously deployed early models of Anthropic for specific operations. Pentagon’s designation does not completely cut off cooperation, but it highlights the core contradiction in AI governance: companies want to set ethical boundaries, while governments emphasize flexibility under national security needs.

This context lends sensitivity to Mythos meetings. On the one hand, the government warns banks to guard against AI-driven attacks, but on the other hand, it has legal disputes with model developers, reflecting the reality that the regulatory framework lags behind technological development.

4. Potential impact on the global financial system

The financial system is highly dependent on digital infrastructure, from payment clearing to trading platforms to the core systems of the Federal Reserve and commercial banks. If the Mythos class model is used for malicious purposes, long-term latent vulnerabilities may be discovered and exploited to achieve remote code execution, privilege escalation, or data manipulation. Imagine: an attacker invades the bank network through chain vulnerabilities, tampering with transaction records, creating the illusion of liquidity, or directly paralyzing the clearing system.

Systemically important banks (G-SIBs) are the backbone of global financial stability. Once its stability is damaged, it may trigger a chain reaction: runs, credit freezes, cross-border payment interruptions, and even amplify into a systemic crisis. The emergency meeting of the Federal Reserve and the Treasury Department is precisely to prevent such "black swan" events. Powell previously emphasized that cyber threats are the primary risk to financial stability, and this action reflects the forward-looking nature of supervision.

At a broader level, the improvement of AI network capabilities will reshape the threat landscape. Traditional defenses rely on human experts and automated tools, while autonomous AI attackers can operate 24/7 and quickly iterate strategies. Critical infrastructure (such as power, communications) and financial systems are highly interconnected, and a single breakthrough may spread across the world. Experts point out that the proliferation of such tools may significantly reduce the cost of attacks and extend the threshold from state actors to technology-savvy non-state actors.

Current data shows that the frequency of cyber incidents has increased in 2025-2026, and the number of AI-assisted attacks has increased. Although Mythos has a limited release, the rapid iteration of similar models means that the risk window is shrinking. Banks need to accelerate the adoption of AI-driven defense while strengthening supply chain security and zero-trust architecture.

5. The strategic positioning of precious metals in the era of uncertainty

Amid technology-driven uncertainty, gold and silver highlight their unique advantages as traditional stores of value. They do not rely on any digital system, require no third-party trust, and cannot be manipulated by code "hacking". In essence, precious metals are physically existing assets whose value is derived from scarcity, historical recognition, and dual industrial/monetary attributes.

Gold has long been a safe-haven asset and has performed stably amid geopolitical tensions, inflationary pressures or systemic financial risks. In 2025-2026, gold prices have been supported by multiple factors, including continued central bank purchases, the trend of de-dollarization, and doubts about the traditional monetary system. Despite short-term fluctuations, consensus forecasts show that gold will remain strong in 2026, and some institutions have raised their long-term expectations.

Silver has both monetary and industrial properties. Industrial demand (solar energy, electric vehicles, AI data center electronic components) continues to grow, driving structural shortages. Silver price forecasts have been revised up significantly in 2026, with some analysts suggesting that industrial use will dominate pricing while its safe-haven role provides additional support amid financial uncertainty. Although the expansion of AI infrastructure increases silver consumption, it also indirectly amplifies the demand for hedging under geopolitical and technological risks.

Precious metals are "simple and reliable" compared to AI-driven digital fragility. They are immune to cyberattacks and cannot be remotely tampered with or created. In a worst-case scenario—a brief disruption to the financial system or a breakdown in trust—physical precious metals can serve as a direct medium of exchange or a store of wealth. Historical experience shows that in times of crisis, precious metals often regain recognition.

However, precious metals are not everything. Prices are affected by macro factors, including U.S. dollar movements, interest rates, inflation expectations and the industrial cycle. Investors should consider diversification, storage security and liquidity. At the same time, technological advancements may introduce new risks but also provide tools for precious metal mining and verification.

6. Risk management and forward-looking suggestions

Faced with AI network threats, response strategies need to be promoted at multiple levels:

Regulation and Governance

Accelerate the formulation of AI dual-use export control and model security assessment standards. Governments and enterprises need to strengthen cooperation and establish a mechanism for sharing threat intelligence.

Technical Defense

Banks and infrastructure operators should invest in AI-assisted defense tools and implement continuous vulnerability scanning, sandbox isolation and human supervision. Zero trust architecture and quantum-safe encryption will be a focus.

Enterprise and individual levels

Decentralize asset allocation and avoid over-reliance on a single number system. Real assets such as precious metals can serve as long-term hedging tools, but they need to be integrated into the overall investment portfolio.

International coordination

AI risks know no borders. Major economies such as the United States, China and Europe need to dialogue on security standards to prevent arms race-style proliferation.

In the long term, AI will reshape the financial and security landscape. The Mythos incident is an early sign that technological acceleration may amplify vulnerabilities. Balancing innovation and security is a common global challenge.

7. Conclusion: Building resilience under the double-edged sword of technology

The emergency meeting triggered by the Anthropic Mythos model highlighted the disruptive potential of AI in the field of cybersecurity. It is both a defensive weapon and a potential offensive weapon. For the global financial system, this is a reminder of the fragility of digital dependence. In times of heightened uncertainty, gold and silver, as “non-smart” assets, offer a path back to fundamental value.

Future development depends on the efforts of many parties: technology companies need to strengthen safety and ethics, regulators need to update their frameworks, and market participants need to improve their resilience. Precious metals will not solve all problems, but in their physical stability and historical role, they embody mankind's pursuit of lasting value. Through careful analysis and diverse preparation, society can better navigate technological advances while reducing systemic risks.