AI hiring filters, harmonised medical device standards, and logs
New York applicants face instant rejections with no visibility into whether AI decided. Meanwhile, medical device AI gets its harmonised standard, and most companies admit they're not ready when AI breaks.
- 01
AI hiring tools in New York may discriminate invisibly
Applicants receive instant rejections with no transparency into whether a human or algorithm decided. The automated screening can perpetuate discrimination around race, sex, age or disability, making unlawful treatment difficult to recognise or challenge.
Why it matters for evidence: Article 86 gives applicants a right to explanation when high-risk AI plays a role in employment decisions, and Article 12 requires those systems to log events.
Source: jdsupra.com
- 02
EN 18286 becomes harmonised standard for medical device AI
The standard was cited in the EU Official Journal in October 2026, providing a conformity route for AI in medical devices under both the AI Act and MDR. It addresses quality management and data governance requirements.
Why it matters for evidence: Article 10 asks for documented data governance, and Article 12 requires automatic logging for high-risk systems, including medical device AI under Annex I from 2 August 2028.
Source: asphalion.com
- 03
82% say they can monitor AI agents, only 25% feel prepared
Executives say their companies can manage AI agents, but only a quarter feel fully prepared when something breaks between those systems. The gap reveals a critical weakness in organisations' ability to demonstrate what their AI did when failures occur.
Why it matters for evidence: Article 12 requires logging of events, and when something breaks the logs are what tell you what happened.
Source: linkedin.com
- 04
Governance platforms should log denied tool calls, not just successes
Composio argues that audit logs must capture denied tool calls to serve as compliance evidence. The article provides evaluation criteria for buyers assessing whether logs meet governance requirements.
Source: composio.dev
- 05
ECB sets 31 October deadline for AI cyber action plans
Traditional model risk frameworks fail to cover most AI systems, while the ECB has set a 31 October 2026 deadline for banks to submit AI cyber action plans. AI governance adequacy has become a measurable supervisory expectation.
Why it matters for evidence: DORA requires financial institutions to protect logs against tampering, deletion and unauthorised access, and Article 12 applies to high-risk AI in finance from 2 December 2027.
Source: aigovernance.com
- 06
Human review requirements for AI underwriting in insurance
AI for life and health insurance underwriting is high-risk under the AI Act, requiring human oversight designed to understand system limitations, monitor operation and override outputs. The article examines criteria that should drive review tier design.
Source: trussed.ai