Progress on Applying AI to Cybersecurity Remains Slow But Steady

While cybersecurity is often cited as one of the first use cases for artificial intelligence (AI) a new survey of 400 security analysts working in organizations with over 1,000 employees in the U.S. conducted by Osterman Research on behalf of ProtectWise, a provider of a cloud-based application for analyzing network packets, finds there is much work to be done before most cybersecurity AI comes anywhere close to living up to the initial hype.

The survey finds 73 percent of respondents reporting they have implemented security products that incorporate at least some aspect of AI. But over half the respondents (54%) says cybersecurity AI delivers inaccurate results. A full 61 percent of respondents also say they don’t believe AI has yet to stop zero-days and advanced threats. Conversely, the survey suggests approximately one-half of respondents either see some value in cybersecurity AI or have not yet formed an opinion. But either way, it’s clear there’s plenty of room for improvement surrounding what is still fairly described as bleeding edge technology.

In fact, a total of 42 percent of the survey respondents say they find cybersecurity AI products as difficult to use, while 46 percent say they find the rules creation and implementation process attached to cybersecurity AI technologies to be burdensome.

Nearly three quarters - 71 percent - also note AI-based products are more expensive than existing traditional cybersecurity technologies and a quarter of respondents (25%) says they do not plan to implement additional AI-enabled security solutions.

Not surprisingly, the survey finds investments in cybersecurity AI technologies are being driven by senior managers rather than cybersecurity professionals. Over half the survey respondents (55%) suggested that the strongest advocates for AI-based security products in their organization are IT executives, while 38 percent identified non-IT executives as being the biggest internal champion. Faced with a chronic shortage of cybersecurity staff, business and IT leaders are clearly hoping AI will at the very least augment whatever cybersecurity expertise they can afford to retain. That faith in cybersecurity AI is at the heart of forecasts that value the market for cybersecurity AI technologies at over $18 billion by 2023.

The real issue with AI is that it’s only as good as the amount of accurate data that gets exposed to the machine and deep learning algorithms that drive an AI model. It’s exceptionally difficult for a startup company that has purported to develop an AI-based cybersecurity offering to gain access to enough data to drive a highly accurate AI model. It’s even a significant challenge for even cybersecurity vendors that have access to large amounts of data to continually update AI models as the nature of the cybersecurity threats organizations face continue to evolve. Because of these issues, most cybersecurity AI technologies need to be delivered as a cloud service. Otherwise, the effort involved in collecting all the data required to drive the AI models would be simply cost prohibitive.

Barracuda Sentinel detects threats that email security gateways can't

A.I.-Based Protection from Spear Phishing, Account Takeover, and Business Email Compromise

While it may be easy to be dismissive of cybersecurity AI at this point, it’s now more of a matter of time before AI has a profound impact on cybersecurity. There will be significant advances in the coming year as the cost of developing AI models that will become increasingly more accurate rapidly declines. That doesn’t mean cybersecurity professionals will be out of a job anytime soon. But it does mean that over the next 12 to 24 months the fundamental nature of those jobs will inexorably change hopefully for the better.