IBM report sees deep fakes emerging as top AI attack threat

IBM study finds deepfake attacks now account for nearly half of AI-enabled breaches as organizations grapple with rising costs and expanding cyber risks.

Key takeaways

• Deepfakes have emerged as the leading AI-enabled attack vector, accounting for 47% of AI-related breaches in the IBM study. 
• AI-enabled breaches are more expensive than average breaches, costing organizations about $6 million on average compared to the global average breach cost of $4.99 million. 
• Organizations are increasing cybersecurity investments in response to AI threats, with spending focused on areas such as alert triage, penetration testing and vulnerability management.

A report published this week by IBM suggests cybercriminals are now mainly embracing artificial intelligence (AI) to launch deep fake attacks that make use of synthetic audio, video, or images of trusted individuals to manipulate unsuspecting end users.

Based on a study of 602 organizations conducted by the Ponemon Institute that reveals one in four malicious breaches (25%) were enabled by artificial intelligence (AI), a 56% increase year over year. Most of the AI-enabled attacks came in the form of deepfake impersonation (47%) and AI-enabled malware (18%). Cybercriminals most often targeted personally identifiable information (PII) of customers more than other types of data by a wide margin. costing $192 per record on average.

The report also notes the average cost of an AI-related breach is on average $6 million, roughly $1 million more than the global breach average of $4.99 million. The average cost of a breach in the U.S., however, is significantly higher at $11.5 million. Those estimates include lost business in addition to the detection and any escalation efforts that might have been required, which collectively account for 63% of the total cost of a breach.

Mean time to identify (MTTI) and contain (MTTC) a breach, meanwhile, is 247 days, a 2.5% increase year over year. More than half of organizations (53%) at the time of the breach also noted that data encryption had been enabled, according to the report.

Additionally, 21% of organizations experienced a breach targeting AI models or applications. The most common causes were weaknesses in compromised application programming interfaces (APIs), applications, or plug-ins (27%) and cloud misconfigurations affecting AI workloads (27%). Among organizations that experienced an AI-related breach, the vast majority (92%) lacked proper AI access controls.

On the plus side, more organizations appear to be now coming to terms with the new realities of the AI arms race cybersecurity teams are now caught up in. A full 85% of organizations plan to increase security spending after becoming aware of the cyber capabilities of advanced AI models. Specific areas of investment include alert triage (60%), penetration testing (52%) and vulnerability scanning and management (37%). However, while 50% of breached organizations report they deployed agents in threat hunting, response and containment, only 18% have thus far applied them to vulnerability scanning and management.

The challenge is that even as organizations invest more broadly in AI, cybersecurity isn’t always at the top of the use cases being considered. AI is proving to be expensive to implement. As a result, organizations are starting to narrow the number of AI initiatives they are willing to fully fund. Cybersecurity leaders are often told to fund AI initiatives by reallocating existing resources. In effect, they are being given the task of rationalizing the existing portfolio of tools and platforms to cover the cost of any AI tool or platform they might want to add. As such, many cybersecurity teams are hoping that rather than replacing tools and platforms they already know, incumbent vendors at a lower total cost and with less disruption will provide them with the AI capabilities they need to thwart attacks that are increasing in both volume and sophistication.

Regardless of approach, the one thing that is clear is time as usual is not on the side of cybersecurity professionals. At this point, it’s not so much a question of whether there will be more incidents in the AI era, but rather how quickly they can be contained using existing tools that hopefully will sooner than later become much more augmented by AI.