AI Card Bao 173.208.162.207

Industry News

This week's Bank of America outage was not due to a hack: AI dug out too many vulnerabilities at once, and patches were rushed too quickly, causing the system to crash.

Several U.S. banks experienced outages this week and customers were unable to log into their accounts. At first many people thought it was a cyber attack.

The answer is even more dumbfounding. Wall Street executives and the National Security Agency explained to members of Congress that the outage was not due to hacking, but because the AI ​​model discovered too many vulnerabilities, and companies rushed to push out patches, causing the system to crash. The shutdown of a classified network of the National Security Agency was also caused by the same reason.

The logic behind it is easy to understand.

Large companies and government agencies now use the most powerful models to scan their systems to find weaknesses before hackers do. Once you find it, it's a race, because attackers have similar tools, or will soon. An insider's original words were to the effect that everyone shoehorned in a large number of patches at the same time.

The normal patching process is rhythmic: evaluation, testing, grayscale, and online. The banking system is particularly conservative, with a lot of old code and complex dependencies. If a patch corrects one mistake, the chain reaction can take away the entire login service. When vulnerabilities come in at ten times the rate of the past, this rhythm is disrupted.

In the same week, a researcher used Codex to dig out a high-risk vulnerability in Java in one hour, and the AI ​​cost was less than $200. Looking at the two things together, the situation is clear: finding loopholes has been accelerated by AI, and fixing loopholes still relies on people and processes.

This is an easily overlooked risk in AI security. Everyone was watching to see if the model would be attacked, but they didn't expect that the defender could cause downtime if he used too much force.

The next competition is probably not who finds the most vulnerabilities, but who can fix them the fastest without crashing the system.

Related