category
AI Risks
Common risks that can happen when AI systems are built, deployed, or used.
Showing 841-860 of 1686 records
Benchmarking (Cross-lingual data contamination) is an AI risk in 6. Socioeconomic and Environmental focused on 6.5 > Governance failure. It is most relevant...
Benchmarking (Raw data contamination) is an AI risk in 6. Socioeconomic and Environmental focused on 6.5 > Governance failure. It is most relevant during 1 -...
Benchmarking (Benchmark leakage or data contamination) is an AI risk in 6. Socioeconomic and Environmental focused on 6.5 > Governance failure. It is most re...
General Evaluations (AI outputs for which evaluation is too difficult for humans) is an AI risk in 7. AI System Safety, Failures, & Limitations focused on 7...
General Evaluations (Biased evaluations of encoded human values) is an AI risk in 6. Socioeconomic and Environmental focused on 6.5 > Governance failure. It...
General Evaluations (Inaccurate measurement of model encoded human values) is an AI risk in 7. AI System Safety, Failures, & Limitations focused on 7.1 > AI...
General Evaluations (Self-preference bias in AI models) is an AI risk in 7. AI System Safety, Failures, & Limitations focused on 7.1 > AI pursuing its own go...
General Evaluations (Difficulty of identification and measurement of capabilities) is an AI risk in 7. AI System Safety, Failures, & Limitations focused on 7...
General Evaluations (Incorrect outputs of GPAI evaluating other AI models) is an AI risk in 7. AI System Safety, Failures, & Limitations focused on 7.1 > AI...
Model Evaluations is an AI risk focused on X.1 > Excluded. It is most relevant during 1 - Pre-deployment.
Fine-tuning related (Catastrophic forgetting due to continual instruction fine-tuning) is an AI risk in 7. AI System Safety, Failures, & Limitations focused...
Fine-tuning related (Degrading safety training due to benign fine-tuning) is an AI risk in 7. AI System Safety, Failures, & Limitations focused on 7.0 > AI s...
Fine-tuning related (Harmful fine-tuning of open-weights models) is an AI risk in 4. Malicious Actors & Misuse focused on 4.2 > Cyberattacks, weapon developm...
Fine-tuning related (Ease of reconfiguring GPAI models) is an AI risk in 4. Malicious Actors & Misuse focused on 4.0 > Malicious use. It is most relevant dur...
Training-related (Poor model confidence calibration) is an AI risk in 7. AI System Safety, Failures, & Limitations focused on 7.3 > Lack of capability or rob...
Training-related (Robust overfitting in adversarial training) is an AI risk in 7. AI System Safety, Failures, & Limitations focused on 7.3 > Lack of capabili...
Model Development is an AI risk focused on X.1 > Excluded. It is most relevant during 1 - Pre-deployment.
Direct Harm Domains (legal and rights-related harms) is an AI risk focused on X.1 > Excluded. It is most relevant during 4 - Not coded.
Direct Harm Domains (system and operational) is an AI risk focused on X.1 > Excluded. It is most relevant during 4 - Not coded.
Post-deployment is an AI risk focused on X.1 > Excluded. It is most relevant during 2 - Post-deployment.