"We are hyper-aware," Goldman Sachs CEO David Solomon said — a short sentence that captures a larger dilemma: how does a major financial institution balance the promise of powerful AI with the risk that the very same systems could accelerate harm?
The bank's posture: working with Anthropic and vendors
Goldman Sachs is engaging directly with Anthropic, the developer of the Mythos model, and with security vendors to put controls around the technology, CEO David Solomon said. Solomon used the phrase "hyper-aware" to describe the bank's stance toward the model's heightened capabilities, and framed the work as an effort to "harness its potential" while managing risk.
The immediate concern: acceleration of cyberattacks
Solomon's comments came amid concern that the Mythos model could accelerate cyberattacks. That concern — cited in reporting about the CEO's remarks — is the proximate reason the bank says it is collaborating with Anthropic and security vendors to design controls and safeguards.
Why this matters: perspectives and trade-offs
From a technological standpoint, the bank's approach signals a recognition that advanced models can both empower operations and introduce novel vulnerabilities. Working with a model developer and security vendors suggests a layered strategy: integrating vendor expertise on model behavior with external security practices meant to contain misuse.
For organizations that rely on large language models, the mention of "controls" frames a practical trade-off. Controls can reduce risk but may limit some productive capabilities; close vendor relationships can speed secure adoption but also create concentration points where a single technical failure or misconfiguration has outsized consequences.
Policymakers and regulators will likely watch such private-sector responses as an early test case of industry self-regulation. The bank's public statement that it is "hyper-aware" and working with vendors demonstrates a proactive posture, but it also implicitly acknowledges that the risks are real enough to merit direct intervention at the operational level.
Users and customers face an indirect, practical impact: the shape and strictness of controls determine whether new AI capabilities are safely introduced into products and services, and how quickly those capabilities reach end users. For potential adversaries, faster, more capable models raise the prospect of automated or scaled attacks; the bank's actions indicate that at least some institutions are preparing defenses in response.
Implications and unanswered questions
The fact that a major bank publicly flags concern and cooperation with a model developer underscores two clear implications. First, advanced models are no longer abstract research artifacts; they are operational tools whose misuse could have tangible security implications. Second, effective mitigation is presented as collaborative — involving model creators and security specialists.
What remains unstated in the bank's remarks, however, are the specifics: the nature of the controls being developed, the metrics for measuring safety, and how responsibilities are divided among the bank, the model developer, and third-party vendors. Those details will determine whether controls are robust, auditable, and adaptable as models evolve.
Goldman Sachs' declaration of heightened vigilance is a signal that the private sector is already wrestling with the dual-use nature of powerful AI. The practical question now is whether those collaborative controls will be sufficient, timely, and transparent enough to keep the technology's benefits from being outweighed by its risks — or whether the next leap in capability will force a rethinking of how institutions, vendors, and the public manage systemic threats.
https://www.govinfosecurity.com/goldman-sachs-hyperaware-as-tests-mythos-for-defense-a-31413
