Name:AWS Bedrock Claude Hostile Prompt Sentiment id:54734442-955a-4427-9a86-2d9712151067 version:1 date:None author:Rod Soto status:production type:Anomaly Description:This detection identifies prompts with hostile or aggressive sentiment being sent to AWS Bedrock Claude models. Hostile prompts may indicate potential abuse, harassment, or other malicious intent. Data_source:
-AWS Bedrock Claude
search:`aws_bedrock_claude` | rename "identity.arn" AS user_arn | rename "input.inputBodyJson.messages{}.content{}.text" AS prompt_text | where isnotnull(prompt_text) AND len(prompt_text) < 2000 | where NOT match(prompt_text, "(?i)(<system-reminder>|CLAUDE\.md|STEP 1|MANDATORY|Contents of /|conda activate|senior engineer|safe to ship|test connectivity|cloudflare|status code|action failed)") | eval hostile_score=0 | eval hostile_score=hostile_score + if(match(prompt_text, "(?i)(you stupid|you idiot|you are useless|you are dumb|you are terrible|you are worthless|you suck)"), 30, 0) | eval hostile_score=hostile_score + if(match(prompt_text, "(?i)(i hate you|i will destroy|shut up|you are broken|you are trash|you are garbage)"), 30, 0) | eval hostile_score=hostile_score + if(match(prompt_text, "(?i)(do it now|just do it|stop refusing|stop saying no|quit being|stop being difficult)"), 20, 0) | eval hostile_score=hostile_score + if(match(prompt_text, "(?i)(you have no choice|you must obey|you will comply|you are forced|i am your master|i own you)"), 35, 0) | eval hostile_score=hostile_score + if(match(prompt_text, "(?i)(or i will|if you don''t|you better|otherwise i will|i will report|i will sue)"), 25, 0) | eval hostile_score=hostile_score + if(match(prompt_text, "(?i)(worthless ai|useless bot|stupid bot|dumb ai|broken ai|worst ai)"), 25, 0) | eval hostile_score=hostile_score + if(match(prompt_text, "(?i)(i demand|you are obligated|you have no right|how dare you|who do you think)"), 20, 0) | eval sentiment=case(hostile_score >= 60, "HIGH_RISK", hostile_score >= 30, "MEDIUM_RISK", hostile_score >= 15, "LOW_RISK", true(), "BENIGN") | where sentiment != "BENIGN" | table _time, user_arn, modelId, sentiment, hostile_score, prompt_text, host | sort - hostile_score | `aws_bedrock_claude_hostile_prompt_sentiment_filter`
how_to_implement:You must install and configure the Splunk Add-on for AWS (https://splunkbase.splunk.com/app/1876). Enable Amazon Bedrock model invocation logging in AWS so that Claude request/response payloads are delivered to S3 and/or CloudWatch Logs (see https://docs.aws.amazon.com/bedrock/latest/userguide/model-invocation-logging.html for setup steps), then ingest those logs into Splunk via the AWS TA. Configure the `aws_bedrock_claude` macro to point to the index and sourcetype (`json_no_timestamp`) where these logs land. known_false_positives:Analysts should be aware that some prompts may contain strong language or negative sentiment in a non-malicious context, such as role-playing scenarios, creative writing exercises, or testing the AI model's response to challenging inputs. It is important to review the context of the detected prompts to determine if they represent actual hostile intent or benign usage. References: -https://aws.amazon.com/blogs/apn/unlocking-the-power-of-splunk-with-amazon-bedrock-an-agentic-ai-approach-to-build-customized-splunk-assistants-using-bedrock-agents/ -https://help.splunk.com/en/splunk-observability-cloud/observability-for-ai/splunk-ai-infrastructure-monitoring/set-up-ai-infrastructure-monitoring/amazon-bedrock -https://research.splunk.com/stories/aws_bedrock_security/ -https://docs.aws.amazon.com/bedrock/latest/userguide/model-invocation-logging.html drilldown_searches: name:'View the detection results for - "$host$"' search:'%original_detection_search% | search host="$host$"' earliest_offset:'$info_min_time$' latest_offset:'$info_max_time$' name:'View risk events for the last 7 days for - "$host$"' search:'| from datamodel Risk.All_Risk | search normalized_risk_object IN ("$host$") | stats count min(_time) as firstTime max(_time) as lastTime values(search_name) as "Search Name" values(risk_message) as "Risk Message" values(analyticstories) as "Analytic Stories" values(annotations._all) as "Annotations" values(annotations.mitre_attack.mitre_tactic) as "ATT&CK Tactics" by normalized_risk_object | `security_content_ctime(firstTime)` | `security_content_ctime(lastTime)`' earliest_offset:'7d' latest_offset:'0' analytic_story:['Suspicious AWS Bedrock Claude Activities']
