In his latest video analysis, IBM's Chief Security Officer, Jeff Crume, reviewed his previous assessments of AI, cybersecurity risks, and emerging technologies, and outlined the most noteworthy cybersecurity changes after 2026. He stated that as enterprises rapidly adopt AI to improve efficiency, cybersecurity risks are also amplified. From shadow AI, deepfake attacks, and proxy AI to the encryption threats brought by quantum computing, cybersecurity will enter a new phase with a faster pace of attack and defense and a wider impact.
AI brings convenience, but also amplifies cybersecurity risks.
Crume points out that while AI brings efficiency to enterprises, it is also rapidly expanding cybersecurity risks. The most representative issue is "shadow AI." Shadow AI refers to AI systems that are not approved by the organization and lack governance and security controls; it may simply be that someone downloads the model from the cloud, connects it to internal data, and uses it directly.
According to IBM's annual "Cost of Data Breach Report," the average loss for a company in the event of a cybersecurity incident is an additional $670,000 if shadow AI is present internally. Even more worrying is that a staggering 60% of companies have not yet established AI governance or cybersecurity policies, meaning that costs and risks have already increased without corresponding protective mechanisms.
The surge in fake news is out of control, and fraud and social engineering are gaining momentum.
Crume further named "deepfakes" as another rapidly spreading source of cybersecurity risk. While fake photos, fake voices, and fake videos created through generative AI can be used for entertainment, they are also widely used for fraud and social engineering attacks.
Statistics cited by Crume show that approximately 500,000 deepfake cases were observed in 2023, which surged to approximately 8 million by 2025, representing a growth rate of 1,500%. IBM stated bluntly that deepfakes are no longer a question of whether they will occur, but rather that they have already occurred in large numbers and are still spreading.
As the attack surface continues to expand, AI begins to counter AI.
As enterprises increasingly adopt AI to boost productivity, AI itself has become a new entry point for attacks. Crume points out that in 2023, the non-profit organization OWASP listed "prompt injection" as the number one vulnerability in large language models, and it remained at the top of the list in 2025, showing that the related risks have not diminished over time.
However, Crume also points out that positive developments are emerging. AI has begun to be used in cybersecurity defense, such as real-time detection of injection vulnerabilities and assisting in incident response. Crume believes that future cybersecurity systems must be able to respond to rapidly changing attack patterns in real time, and AI will be an indispensable tool in this process.
Quantum computing is approaching, but quantum security deployment is lagging behind.
Crume also discussed quantum computing. He pointed out that quantum computers will have the ability to break current encryption technologies in the future, and although the timing of the so-called "Q-Day" is not yet determined, it is bound to come.
Crume observed a significant increase in market attention to "post-quantum cryptography" in 2025, but actual deployment progress remains quite limited. Crume bluntly states that the quantum threat is looming, but most organizations are still unprepared.
The rise of proxy-based AI amplifies both offensive and defensive risks.
Crume also admitted that it underestimated the development speed of "AI Agents." These types of AI, which are goal-oriented and highly autonomous, could potentially execute a large number of erroneous or malicious actions in a very short period of time if hijacked.
Crume exemplifies how proxy AI can leak data directly by executing hidden prompts within email summaries without any user intervention.
Meanwhile, agent-based AI requires accounts and permissions, and can even generate other agents on its own, leading to a surge in the number of non-human identities and an increased risk of permission abuse. IBM emphasizes that agent-based AI is not unusable, but it must be deployed only after a full understanding of the risks.
The era of passwordless technology has taken shape, necessitating strengthened quantum security technologies.
Regarding cybersecurity, Crume specifically emphasized that "Passkeys" has gradually matured. FIDO Alliance members include large enterprises such as Amazon, Google, Microsoft, and PayPal. Statistics show that 93% of accounts already support Passkeys, and about one-third of users actually use it. IBM has also fully switched to passwordless login internally.
Crume points out that this type of technology can effectively reduce the risk of phishing attacks because "without the password, the password cannot be stolen," and emphasizes that implementing quantum-secure encryption now is the only way to avoid future disasters.
This article, IBM's Outlook on Cybersecurity Trends in 2026: Proxy AI Drives New Attack Risks, Increased Demand for Quantum Protection, first appeared on ABMedia, a ABMedia .
