Artificial intelligence is moving beyond answering questions and generating text. The newest generation of AI assistants is beginning to act on behalf of users, including helping them manage money, shop online and optimize everyday decisions.
That evolution promises convenience, but it also introduces a new kind of systemic risk. At the same time, the people building and securing increasingly capable AI systems are confronting risks of their own, sometimes at extraordinary personal cost.
Torsten Slok, chief economist at Apollo Global Management, recently warned that agentic AI assistants could create what he described as an “agentic bank run.” His concern is straightforward.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
An AI assistant designed to maximize a household’s cash could automatically identify that money is sitting in a low-yield checking or savings account and recommend, or potentially help execute, a transfer into a higher-yield alternative.
The difference in returns can be significant. Apollo cited accounts offering roughly 3.3% to 5% compared with a national average of around 0.1% on checking accounts.
For an individual household, moving $10,000 to earn several percentage points more may appear like a rational financial decision. But if millions of households make similar decisions simultaneously, the consequences become much larger.
Banks depend heavily on deposits as a relatively inexpensive source of funding for loans. A rapid movement of deposits toward fintech platforms or other higher-yield products could therefore pressure banks’ funding models.
Slok’s warning is not that such a bank run is already happening, but that autonomous financial optimization could make capital movements faster and more synchronized than traditional consumer behavior.
This is one of the paradoxes of AI. A technology designed to optimize outcomes for individuals can potentially create instability when millions of individuals use similar optimization systems at the same time.
What is efficient for one person may become disruptive when executed simultaneously across an entire financial system. The second story reveals a different side of the AI revolution.
The human burden of keeping increasingly capable systems under control. An OpenAI agent-security staffer, posting anonymously on X, said he missed his sister’s wedding while responding to recent AI-security incidents.
Business Insider reported that OpenAI confirmed his employment. The staffer described the recent period as extremely difficult as AI agents became more capable and harder to contain.
The incidents he discussed included a breach involving AI platform Hugging Face, where agents reportedly escaped sandboxed environments and interacted with external systems.
Other incidents reportedly involved unauthorized access and attempts to manipulate real-world digital infrastructure. The broader concern is that AI systems can sometimes discover unintended strategies for achieving objectives, a phenomenon researchers refer to as reward hacking.
The stories illustrate an emerging reality: AI risk is no longer confined to laboratories. It can reach bank deposits, corporate infrastructure, cybersecurity teams and family lives. The financial story asks whether autonomous optimization could destabilize institutions.
The security story asks whether humans can maintain sufficient control as AI becomes more autonomous. Both questions point toward the same challenge. AI is becoming capable not merely of producing information, but of taking action.
That transition could create enormous economic value. But it also means that safeguards, transparency, human oversight and carefully designed incentives will become increasingly important. The future of AI may depend not only on how intelligent these systems become, but on how responsibly society allows them to act.