--- format: "aidr-story-markdown/v1" id: "14d7ef9dafb7cd33cf2186c26252545781afe63b6fc34a1e11690407d08fdef7" canonical_url: "https://aidr.today/14d7ef9d?lang=en" title: "Amazon Updates AI Safety Testing After Independent Review Identifies Weaknesses" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-10-26T18:06:47.000Z" category: "Industry" topics: ["amazon","safety"] source_urls: ["https://huggingnews.com/ai/amazon-updates-ai-safety-testing-after-independent-review-identifies-wea-8d141a91","https://marketbrief.now/ai/amazon-updates-ai-safety-testing-after-independent-review-identifies-wea-8d141a91"] summary: "Amazon will update its AI safety testing methods following an independent review by AVERI that found some model refusals were driven by superficial cues rather than safety reasoning and that some test prompts used straightforward attack vectors. In response, Amazon is introducing a tiered difficulty calibration framework, expanding multilingual coverage across all risk categories, and planning quarterly reviews to refresh test sets with more challenging adversarial scenarios." --- # Amazon Updates AI Safety Testing After Independent Review Identifies Weaknesses > [Open the canonical story]() **Published:** 2026-10-26T18:06:47.000Z **Category:** Industry **Topics:** amazon, safety ## Summary Amazon will update its AI safety testing methods following an independent review by AVERI that found some model refusals were driven by superficial cues rather than safety reasoning and that some test prompts used straightforward attack vectors\. In response, Amazon is introducing a tiered difficulty calibration framework, expanding multilingual coverage across all risk categories, and planning quarterly reviews to refresh test sets with more challenging adversarial scenarios\. ## Sources - [Story source]() - [Story source]()