Exam-style
A marketing team wants to publish AI-generated images and copy on the company's public website, replacing a photo shoot and a freelance copywriter. Legal will not sign it off. They expect a third party to turn up later claiming a published image or paragraph reproduces work the company had no right to use, and to ask who pays. Which set of measures addresses that risk?
Reveal the answer
C. Choose a model whose provider indemnifies generated output, read its licence and acceptable use policy before selecting it, ground generation in content the company already owns, and have a person review every asset before it publishes
Legal has described an intellectual property infringement claim, one of the legal risks of working with generative AI. A third party asserts that published output reproduces their protected work, and the argument is over liability. Guardrails content filters block hate, insults, sexual content, violence and misconduct; an inoffensive paragraph can still be somebody else’s paragraph. A contextual grounding check scores a response against its source, which addresses veracity, and a faithfully grounded sentence still infringes if the source was not the company’s to use. KMS customer managed keys keep material confidential, which has no bearing on content published on purpose. Titan Image Generator watermarks every image it generates, as Nova Canvas did before its end of life, and Bedrock’s detection API, in preview, confirms as much, which settles provenance rather than ownership. Only C shifts liability. AWS offers uncapped indemnity for copyright claims on output from the generally available services its Service Terms list, conditional on responsible use such as leaving the filters on. Licence and acceptable use terms set what the output may be used for commercially, so read them before selecting.
Q. Marketing wants generated images and copy on the public site. Legal is worried a third party will claim the output reproduces their protected work. Which controls answer that?
A. Pick a model whose provider indemnifies generated output, read its licence and acceptable use policy before you select it, ground generation in content you already own, and keep a person reviewing assets before they publish.
Why? Legal has described intellectual property infringement claims, one of the legal risks of working with generative AI; the argument is over who pays. Contractual indemnity is the only control that moves liability. AWS covers copyright claims on output from the services its Service Terms list as indemnified, provided you have not fed in infringing material or turned the filters off. Licensing sets what you may do with the output commercially, and it differs sharply between hosted proprietary models and open weights, so read it during the model choice itself. The other four close different risks. Guardrails content filters cover safety, a contextual grounding check covers veracity, KMS encryption covers confidentiality, and watermarking with its detection API, still in preview, covers provenance. A faithfully grounded, well-encrypted, watermarked paragraph can still be somebody else’s paragraph. Human review catches end user risk: a person is the last chance to stop a generated claim somebody might act on.