Artificial intelligence monsters move from film to reality

For more than 100 years, literary minds and filmmakers have created vivid and terrifying stories about technology gone awry. Mary Shelley’s 1918 novel Frankenstein depicts a creature made from human cadavers who is brought to life only to go on a murderous rampage killing his creator and many others. 2001: A Space Odyssey, the Stanley Kubrick 1968 film, depicts HAL 9000, a computer with a human personality and creepy voice, who kills the crew of a spaceship after it believes humans are attempting to interfere with its mission.

In June, at the annual Royal Aeronautical Society summit in London, Colonel Tucker “Cinco” Hamilton described a war simulation in which a drone controlled by artificial intelligence decided to “kill” its human operator for obstructing its mission objectives. The drone was programmed to destroy a surface-to-air missile site, but the final decision whether to strike was to be made by a human, according to Hamilton, who is the head of the US Air Force’s AI Test and Operations.

Hamilton is reported as saying that over time the drone determined that it was not earning “points” for bombing the target because a human controller prevented it. So the drone variously “killed” the human controller or knocked out control towers used by the human to communicate with the drone. The U.S. Air Force has since denied Hamilton’s account, but the scenario certainly sounds plausible and is likely to happen unless AI programming includes an ethical decision tree that a human would presumably engage prior to making a go/no-go decision.

AI has also become a monster weapon for cyber criminals. In its 10th annual Data Breach Industry Forecast, Experian notes that AI has significantly improved cyber criminals’ ability to launch attacks. The days of phishing emails full of typos, bad grammar and unbelievable claims of riches to be inherited from an unknown uncle who just died are gone. Cyber criminals now use AI to collect personal data and create text that is much more difficult to detect as fake. For example, a recent phishing email is disguised as a notice that the license for Microsoft software is about to expire and provides a convenient button to “renew” which actually puts malware on the user’s computer.

The Experian report also notes that cyber criminals use AI to find vulnerabilities in security and once they gain access to an organization’s network, they “play the long game” to avoid raising suspicion. AI monitors the network and makes a series of small inroads that are not egregious enough to throw up security red flags. The criminals then infiltrate a company’s network gaining access to more and more desired data over time. AI may even enable cyber criminals to carry out a very sophisticated cat and mouse game in which hackers feed bad information to a security system to misdirect it to search for malware that won’t actually be used in an attack, clearing the way for the real malware to operate.

This more sophisticated AI-enabled criminal is increasingly targeting professional service companies such as law firms, not only because they have large amounts of sensitive client data, but also because they provide a window into ongoing transactions. Experian notes that cyber criminals will hack into email communications and monitor a developing transaction, then use that information to create very credible phishing emails or otherwise masquerade as a participant to the transaction at just the right time to misdirect payments or sensitive information.

The longer an attack persists, the more expensive it may be. The Experian report cites data that companies that contain a breach in less than 30 days save more than $1 million compared to breaches that remain undetected for longer. An IBM report reveals that on average, it takes 212 days for organizations to identify a cyber attack and 75 days to contain it. RiskIQ calculates that cybercrime cumulatively costs organizations $2.9 million every minute.

Consistent with the long-game approach, cyber criminals have modified ransomware attacks. Previously a hacker would infiltrate a system and shut it down until a ransom was paid. Increasingly, hackers leave the system functional, but threaten to disclose the entity’s information publicly or sell it to the dark web unless ransom is paid. Experian reports that ransomware attacks increased 414% in the first six months of this year. One driver for this increase is that enterprising cyber criminals have developed so-called ransomware-as-a-service, which essentially “licenses” ransomware tools to other cyber criminals. The developer of the ransomware may be paid a set fee, or may get a percentage of ransom payment extracted by hackers successfully using the tool.

While the number and increasing sophistication of cyber criminals may seem overwhelming, AI is also a powerful weapon to fight cybercrime. Law enforcement agencies are partnering with cyber security experts and have developed solutions, including decryption tools that help victims recover data without having to pay a ransom. It seems clear that it will take global cooperation to combat global cyber criminals, and such efforts should be made a priority.