ACM Policy Committees

1/ Removing names from a dataset doesn’t make it anonymous. That technical question now sits at the center of a proposed $10M sale of Spirit Airlines’ internal data to Google for AI training. 🧵 www.pymnts.com/big-data/202...

4/ Their requests: exclude employee data where possible, build de-identification with worker input, keep heightened protections for medical and disciplinary records, and bar using the data to profile groups of workers.

1/ Removing names from a dataset doesn’t make it anonymous. That technical question now sits at the center of a proposed $10M sale of Spirit Airlines’ internal data to Google for AI training. 🧵 www.pymnts.com/big-data/202...

2/ Per a letter from 121 members of Congress, court findings show the data would include ~100M emails, 500M Microsoft Teams messages, employee records, timecards, payroll and tax info, and employment contracts from former Spirit employees.

ACM’s Expanding Regions Technology Policy Committee has submitted comments to the African Union’s Open Consultation on AI Governance and Regulation in Africa. www.acm.org/public-polic... buff.ly/TVu5vDd

AI can write a patch in seconds. A human maintainer still has to decide whether to accept it. That gap is the open source story of the next five years. New ACM TechBrief 🧵 dl.acm.org/doi/epdf/10....

4/ SBOMs are mandated in the US and EU. Most open source releases still don't ship one. And an SBOM won't tell you whether a component is maintained, funded, or abandoned.

AI can write a patch in seconds. A human maintainer still has to decide whether to accept it. That gap is the open source story of the next five years. New ACM TechBrief 🧵 dl.acm.org/doi/epdf/10....

2/ Firms would spend 3.5x more on software if open source didn't exist. Estimated demand-side value: $8.8 trillion. This is critical infrastructure, and much of it runs on volunteer labor.

The frontier AI competition isn't won by the biggest computers. It's won by the most trustworthy systems. ACM Europe just released a policy brief reframing Europe's AI strategy—and it's a direct challenge to the "arms race" narrative. 🧵

Europe's layered sovereignty strategy: Transparent supply chains for frontier systems AI Factories as federated trusted environments (not just compute) Proportionate (case-by-case) safeguards for open-weight models International coordination on evaluation & incident response

The frontier AI competition isn't won by the biggest computers. It's won by the most trustworthy systems. ACM Europe just released a policy brief reframing Europe's AI strategy—and it's a direct challenge to the "arms race" narrative. 🧵

Most AI governance treats trustworthiness as a compliance layer added after the fact. Wrong move. Instead: Purpose-first design. Ask why you're building a system, whose interests it serves, whether benefits justify risks. Build that into the architecture from day one.

1/ Consumers accept AI system outputs without any further fact-checking more than 90% of the time. The FTC cites that number in its policy statement on accuracy in AI systems. ACM's US Technology Policy Committee agrees the problem is real — and says the diagnosis is wrong.

4/5 And there's no clean line to regulate. In one session, a user may move from fact-seeking prompts to creative ones. No natural delineation exists that's easy to implement.

1/ Consumers accept AI system outputs without any further fact-checking more than 90% of the time. The FTC cites that number in its policy statement on accuracy in AI systems. ACM's US Technology Policy Committee agrees the problem is real — and says the diagnosis is wrong.

2/5 The bigger problem isn't ideology. It's confabulation: systems stating things confidently that aren't true. Grounding methods like retrieval-augmented generation help, but aren't infallible.

1/ Europe is finalizing its cybersecurity standard for AI systems (prEN 18282). ACM's Europe Technology Policy Committee reviewed the draft and recommends 6 improvements. Thread 🧵

4/ Agentic AI needs clear division of labor: providers design safeguards, deployers configure & monitor them. Neither should let an AI system change its own safety settings unsupervised.

1/ Europe is finalizing its cybersecurity standard for AI systems (prEN 18282). ACM's Europe Technology Policy Committee reviewed the draft and recommends 6 improvements. Thread 🧵

2/ The draft rightly flags AI-specific attack surfaces: training data, model weights, prompts, connected tools. Conventional ICT security standards miss these.

1/ Africa needs ~700 more data centers. South Africa already hosts the continent's most developed market. ACM just told the South African Human Rights Commission: this is a human rights question, not only an infrastructure one. Thread 🧵

4/ Hosting data in South Africa doesn't guarantee South African legal control. Foreign-owned parent firms can still face outside access demands. Data residency isn't data sovereignty.

1/ Africa needs ~700 more data centers. South Africa already hosts the continent's most developed market. ACM just told the South African Human Rights Commission: this is a human rights question, not only an infrastructure one. Thread 🧵

2/ Data centers draw heavily on scarce energy & water, often for foreign-owned operators' profit. Should public resources subsidize offshore gains? ACM says approvals need proof of local benefit.

1/ Seven computing & data science orgs just told OMB the same thing: withdraw this rule. Here's why they're worried. 🧵

4/ For CS/AI researchers specifically: vague provisions could create legal risk around work on algorithmic fairness and bias mitigation — a core, active research area.

1/ Seven computing & data science orgs just told OMB the same thing: withdraw this rule. Here's why they're worried. 🧵

2/ The proposed rule (OMB-2026-0034) would let a political appointee approve or block discretionary grants — demoting peer review to "advisory" status.