The Easy Way In/Out: Securing The Artificial Future

Other AI-enabled activities to watch out for

As AI becomes more agentive and begins using enterprise tools and computers autonomously, it creates a chain of events and interactions that are invisible to human operators. This lack of visibility can be a security concern as it will be a struggle to monitor and control the agents’ actions in real time.

As enterprises inevitably enter the AI race, they may expose themselves to more vulnerabilities and flaws. Sensitive information disclosure is a growing concern in that LLMs risk exposing sensitive data during interactions with employees and customers, including personal identifiable information and intellectual property.

More companies are using AI to discover infrastructure vulnerabilities, increasing both the number of identified vulnerabilities, and, potentially, the risk of exploitation. The AI agents will also become more attractive to malware authors. There is great potential for unauthorized or malicious activities carried out by misguided autonomous agents, including "agent hijacking" by external actors. Traditional malware and known threat detection will shift left toward vulnerability and attack surface management, while also shifting right towards leveraging foundational data intelligence.

System resource consumption by AI agents, whether benign or malicious, can also lead to denial-of-service when resources are overwhelmed.

Enterprises adopting agentive AI should beware of

Data theft

Data breaches caused by coding errors

Model misalignment

Infrastructure-based DDoS attacks

Compromise from third-party libraries or code

Nation-state threat actors such as Lazarus, Turla, and Pawn Storm were particularly active in 2024 and are expected to increase their activity in 2025. These groups will continue to target diplomatic information and military technologies as well as their supply chains for maximum impact. Other criminal and mercenary groups are now doing a significant amount of espionage too, such as Void Rabisu and the DaVinci Group. Operations aligned with Russian propaganda like Doppelganger and pro-China networks such as Spamouflage are seen to continue leveraging disinformation to deepen societal divisions. Salt Typhoon’s recent attack that siphoned phone call audio and data could suggest that nation-state threat actors are exploring audio deepfakes.

Earth Hundun and other nation-state threat actors with similar geopolitical alignments will continue their speedy evolution. Meanwhile, North Korean groups will likely keep focusing on cryptocurrency to help bypass sanctions. APT29 has been targeting cloud environments, an activity expected to rise. Sandworm, primarily involved in operations related to the invasion of Ukraine, could expand its operations depending on future geopolitical developments. The same geopolitical developments will continue to see hacktivist groups impacting enterprises they see as linked to their political or ethical ideals; and in the act are influenced by state actors looking to make use of the free energy they provide.

APT groups are expected to persist

Memory management vulnerabilities, such as Out-of-Bounds (OOB) Write/Reads, continue to be exploited by threat actors; the CISA KEV reported 18 different OOB Write vulnerabilities exploited this year. Memory management bugs (CWE-787 and CWE-125) are among the top most dangerous vulnerabilities based on severity and frequency as analyzed by MITRE. Memory management and memory corruption bugs will likely continue to be favorites of attackers in the coming year. Memory corruption and logic bugs are often combined to create winning Pwn2Own entries.

In 2025, we expect to see bug chains continue to be prevalent while bugs in APIs create problems for cloud resources. We also expect more container escapes. We had a Docker escape demonstrated this year, and more are sure to follow.

However, older, simpler vulnerabilities like cross-site scripting (XSS) and SQL injections continue to be popular and will continue to be targeted for as long as they exist. This is especially true for operational technology devices and services that security experts have yet to scrutinize. Older techniques continue to be effective like simple buffer overflows and command injection exploits remain popular for devices like printers and Wi-Fi cameras, as these devices were not built with security in mind. Despite the advancement that AI offers, these classic avenues of least resistance will continue being useful for cybercriminals who want maximum profit for the least amount of effort.

Meanwhile, security is faced with the challenge of keeping pace with mobility innovation as the path forward continues to unfold. The push for platform standardization in mobility across different platforms means that a single vulnerability in a widely adopted system could cascade across multiple models and manufacturers, leading to large-scale recalls or urgent software updates. Vehicle systems are also transitioning from multiple distributed electronic control units (ECUs) to a few centralized ECUs with higher computational power. While this transition reduces the cost and time required for developing new versions of E/E systems, it in turn lowers the barriers to entry, reduces the learning curve for attackers, and increases the risks of widespread attacks across various OEM platforms that share components.

This year, EV charging and in-vehicle infotainment (IVI) systems were frequently exploited at Pwn2Own, especially during the inaugural Pwn2Own Automotive event. These systems represent a broad attack surface that vendors have yet to thoroughly vet for security gaps. In the future, attackers are likely to target the communication between EV chargers and vehicles, as EV charging stations are connected online and interact with payment systems and the power grid as well as vehicles. If not properly secured, they could be exploited to access in-vehicle systems, steal payment information, and disrupt charging processes.

Cyberthreats to connected cars could lead to serious physical consequences. For example, compromised vehicles could enable criminals to disable brakes or steering. Advanced safeguards and proactive measures are critical to protecting automotive systems from cyber-physical threats that endanger driver, passenger, and pedestrian safety.

Risks in mobility game changers

2024 saw a rise in ransomware groups leveraging legitimate tools for data exfiltration, credential collection and replication, which can make it easier for attackers to move laterally and escalate privileges. As we move into 2025, legitimate tools will continue to be exploited as cybercriminals realize their potential in disguising attack activity as legitimate and that they already have access to valuable resources, data, and enterprise networks. With the rise of ransomware attacks starting through vulnerabilities or using compromised accounts, attacks that start with phishing will likely go down, suggesting a shift in techniques for ransomware gangs. More successful attacks from our recent investigations used compromised accounts to connect to a machine in the environment, while other cases saw the attacker bypassing multi-factor authentication (MFA) mechanisms. Ransomware attacks could also drift towards business models that no longer necessitate encryption.

As ransomware groups evolve in their technique of leveraging legitimate tools, organizations should not only rely on malicious files and hash detections but also monitor behavior across layers. Enterprises should opt for solutions that provide enhanced visibility and correlated detection across multiple layers, ensuring that incidents with the potential to cause significant system damage can be addressed as early as possible. Organizations can stay on top of threats by subscribing to CTI platforms to gain insights and information on the tactics, techniques, and procedures of cyberattacks, to prepare prevention and mitigation protocols.

Ransomware attacks will see a rise in the use of

In the past, we have seen a one-to-one ratio between loaders and info stealers, where one loader will install just one infostealer, but we have started seeing multiple infostealers installed by a single loader. Information harvested by such threats are useful for threat actors and enable them to carry out other attacks. Ransomware groups will continue to use this as a key part of their attacks: utilize information, such as user accounts harvested by infostealers, in their ransomware attacks.

Malvertising threats have been thrust into the spotlight partly because of the widespread proliferation of infostealers that use this arrival technique, which could lead to attackers seeing its potential for other campaigns. We have already seen ransomware groups use this to get a foothold in a target environment. The stealth mechanisms described in the previous section show how threat actors can innovate and elevate the technique to make it more effective in gaining initial access. In 2025 we predict this trend will continue.