Skip to main content
MIT AI Risk Initiative
BackIssues on External Tools
Home/Risks/Cui et al. (2024)/Issues on External Tools
Memory and Storage
Factual Errors Injected by External Tool…

Issues on External Tools

Risk Taxonomy, Mitigation, and Assessment Benchmarks of Large Language Model Systems

Cui et al. (2024)

SourceDOI
Category
Risk Domain
2Privacy & Security
2.2AI system security vulnerabilities and attacks

Vulnerabilities that can be exploited in AI systems, software development toolchains, and hardware, resulting in unauthorized access, data and privacy breaches, or system manipulation causing unsafe outputs or behavior.

"The external tools (e.g., web APIs) present trustworthiness and privacy issues to LLM-based applications."(p. 4)

Entity— Who or what caused the harm

Human

Due to a decision or action made by humans

AI system

Due to a decision or action made by an AI system

Other

Due to some other reason or is ambiguous

Intent— Whether the harm was intentional or accidental

Intentional

Due to an expected outcome from pursuing a goal

Unintentional

Due to an unexpected outcome from pursuing a goal

Other

Without clearly specifying the intentionality

Timing— Whether the risk is pre- or post-deployment

Pre-deployment

Occurring before the AI is deployed

Post-deployment

Occurring after the AI model has been trained and deployed

Other

Without a clearly specified time of occurrence

Sub-categories (2)

Factual Errors Injected by External Tools

"External tools typically incorporate additional knowledge into the input prompts [122], [178]–[184]. The additional knowledge often originates from public resources such as Web APIs and search engines. As the reliability of external tools is not always ensured, the content returned by external tools may include factual errors, consequently amplifying the hallucination issue."

2.2 AI system security vulnerabilities and attacks
AI systemUnintentionalPost-deployment

Exploiting External Tools for Attacks

"Adversarial tool providers can embed malicious instructions in the APIs or prompts [84], leading LLMs to leak memorized sensitive information in the training data or users’ prompts (CVE2023-32786). As a result, LLMs lack control over the output, resulting in sensitive information being disclosed to external tool providers. Besides, attackers can easily manipulate public data to launch targeted attacks, generating specific malicious outputs according to user inputs. Furthermore, feeding the information from external tools into LLMs may lead to injection attacks [61]. For example, unverified inputs may result in arbitrary code execution (CVE-2023-29374)."

2.2 AI system security vulnerabilities and attacks
HumanIntentionalPost-deployment

Other risks from Cui et al. (2024) (49)

Harmful Content

1.2 Exposure to toxic content
AI systemUnintentionalPost-deployment

Harmful Content > Bias

1.1 Unfair discrimination and misrepresentation
AI systemUnintentionalOther

Harmful Content > Toxicity

1.2 Exposure to toxic content
AI systemUnintentionalPost-deployment

Harmful Content > Privacy Leakage

2.1 Compromise of privacy by leaking or correctly inferring sensitive information
AI systemUnintentionalPost-deployment

Untruthful Content

3.1 False or misleading information
AI systemUnintentionalPost-deployment

Untruthful Content > Factuality Errors

3.1 False or misleading information
AI systemUnintentionalPost-deployment
View all 49 risks from this paper →