Modify On-Board Values: Poison AI/ML Training Data
Threat actors may perform data poisoning attacks against the training data sets that are being used for artificial intelligence (AI) and/or machine learning (ML). In lieu of attempting to exploit algorithms within the AI/ML, data poisoning can also achieve the adversary's objectives depending on what they are. Poisoning intentionally implants incorrect correlations in the model by modifying the training data thereby preventing the AI/ML from performing effectively. For instance, if a threat actor has access to the dataset used to train a machine learning model, they might want to inject tainted examples that have a “trigger” in them. With the datasets typically used for AI/ML (i.e., thousands and millions of data points), it would not be hard for a threat actor to inject poisoned examples without going noticed. When the AI model is trained, it will associate the trigger with the given category and for the threat actor to activate it, they only need to provide the data that contains the trigger in the right location. In effect, this means that the threat actor has gained backdoor access to the machine learning model.
Other Subtechniques of Modify On-Board Values ( 13 )
When AI/ML is being used for mission critical operations, the integrity of the training data set is imperative. Data poisoning against the training data set can have detrimental effects on the functionality of the AI/ML. Fixing poisoned models is very difficult so model developers need to focus on countermeasures that could either block attack attempts or detect malicious inputs before the training cycle occurs. Regression testing over time, validity checking on data sets, manual analysis, as well as using statistical analysis to find potential injects can help detect anomalies.
Employ the principle of least privilege, allowing only authorized processes which are necessary to accomplish assigned tasks in accordance with system functions. Ideally maintain a separate execution domain for each executing process.
Implement disaster recovery plans that contain procedures for taking regular data backups that can be used to restore critical data. Ensure backups are stored off system and is protected from common methods adversaries may use to gain access and destroy the backups to prevent recovery.
Utilize on-board intrusion detection/prevention system that monitors the mission critical components or systems and audit/logs actions. The IDS/IPS should have the capability to respond to threats (initial access, execution, persistence, evasion, exfiltration, etc.) and it should address signature-based attacks along with dynamic never-before seen attacks using machine learning/adaptive technologies. The IDS/IPS must integrate with traditional fault management to provide a wholistic approach to faults on-board the spacecraft. Spacecraft should select and execute safe countermeasures against cyber-attacks. These countermeasures are a ready supply of options to triage against the specific types of attack and mission priorities. Minimally, the response should ensure vehicle safety and continued operations. Ideally, the goal is to trap the threat, convince the threat that it is successful, and trace and track the attacker — with or without ground support. This would support successful attribution and evolving countermeasures to mitigate the threat in the future. “Safe countermeasures” are those that are compatible with the system’s fault management system to avoid unintended effects or fratricide on the system.
Ensure fault management system cannot be used against the spacecraft. Examples include: safe mode with crypto bypass, orbit correction maneuvers, affecting integrity of telemetry to cause action from ground, or some sort of proximity operation to cause spacecraft to go into safe mode. Understanding the safing procedures and ensuring they do not put the spacecraft in a more vulnerable state is key to building a resilient spacecraft.
Provide the capability to enter the spacecraft into a configuration-controlled and integrity-protected state representing a known, operational cyber-safe state (e.g., cyber-safe mode). Spacecraft should enter a cyber-safe mode when conditions that threaten the platform are detected. Cyber-safe mode is an operating mode of a spacecraft during which all nonessential systems are shut down and the spacecraft is placed in a known good state using validated software and configuration settings. Within cyber-safe mode, authentication and encryption should still be enabled. The spacecraft should be capable of reconstituting firmware and software functions to pre-attack levels to allow for the recovery of functional capabilities. This can be performed by self-healing, or the healing can be aided from the ground. However, the spacecraft needs to have the capability to replan, based on equipment still available after a cyber-attack. The goal is for the spacecraft to resume full mission operations. If not possible, a reduced level of mission capability should be achieved. Cyber-safe mode software/configuration should be stored onboard the spacecraft in memory with hardware-based controls and should not be modifiable.
Identify the key system components or capabilities that require isolation through physical or logical means. Information should not be allowed to flow between partitioned applications unless explicitly permitted by security policy. Isolate mission critical functionality from non-mission critical functionality by means of an isolation boundary (implemented via partitions) that controls access to and protects the integrity of, the hardware, software, and firmware that provides that functionality. Enforce approved authorizations for controlling the flow of information within the spacecraft and between interconnected systems based on the defined security policy that information does not leave the spacecraft boundary unless it is encrypted. Implement boundary protections to separate bus, communications, and payload components supporting their respective functions.