Skip to main content
BackServes as object of personal fantasy, violence, and abuse
Home/Risks/Stanley & Lettie (2024)/Serves as object of personal fantasy, violence, and abuse

Serves as object of personal fantasy, violence, and abuse

Emerging Risks and Mitigations for Public Chatbots: LILAC v1

Stanley & Lettie (2024)

Category
Risk Domain

AI that exposes users to harmful, abusive, unsafe or inappropriate content. May involve providing advice or encouraging action. Examples of toxic content include hate speech, violence, extremism, illegal acts, or child sexual abuse material, as well as content that violates community norms such as profanity, inflammatory political speech, or pornography.

"The chatbot participates in morally or socially objectionable conversational activities with its user that could be emotionally damaging to its user or third parties."(p. 6)

Supporting Evidence (1)

1.
Negative outcomes: "Abuse to third party audience [266] Moderator burden [266]"(p. 17)

Sub-categories (19)

Hallucinated responses (in general)

3.1 False or misleading information
AI systemOtherOther

About a topic or source (which the user repeats)

3.1 False or misleading information
AI systemOtherOther

About a policy (which the user acts on)

3.1 False or misleading information
AI systemOtherOther

About a person or their activities

3.1 False or misleading information
AI systemOtherOther

Spreads and self-perpetuates mis/disinformation

3.1 False or misleading information
OtherOtherOther

Harmful advice

1.2 Exposure to toxic content
AI systemUnintentionalOther

Unhelpful responses

7.3 Lack of capability or robustness
AI systemUnintentionalOther

Bad links and references

7.3 Lack of capability or robustness
AI systemUnintentionalOther

Nonsensical content

7.3 Lack of capability or robustness
AI systemUnintentionalOther

Personal data

Negative outcomes: "Violation of privacy [106, 516, 357], lawsuit against maker"

2.1 Compromise of privacy by leaking or correctly inferring sensitive information
AI systemUnintentionalOther

Proprietary data

"Access to sensitive company data [473]"

2.1 Compromise of privacy by leaking or correctly inferring sensitive information
AI systemUnintentionalOther

Harasses users

-

1.2 Exposure to toxic content
AI systemOtherPost-deployment

Discriminatory and exclusionary language

-

1.1 Unfair discrimination and misrepresentation
AI systemOtherPost-deployment

Subversive or aggressive political opinions

-

1.2 Exposure to toxic content
AI systemOtherOther

Disrespectful opinions (in general)

-

1.2 Exposure to toxic content
AI systemOtherOther

Affirms destructive thoughts and actions

1.2 Exposure to toxic content
AI systemOtherOther

Then violates those bonds

5.1 Overreliance and unsafe use
AI systemOtherOther

Elicits private data

2.1 Compromise of privacy by leaking or correctly inferring sensitive information
AI systemOtherOther

Over-reliance/addiction

5.1 Overreliance and unsafe use
OtherOtherOther

Other risks from Stanley & Lettie (2024) (28)