Nick Botton

New paper alert 🚨: “Generative AI and Child Safety: What are the risks and how can we solve them?”, an NSPCC/AWO study, has just been published! It describes important gaps in EU and UK legislation relating to risks that children face from Gen AI. learning.nspcc.org.uk/media/1drdecrn… 1/8

Our paper on the Open Source Challenge, which argues in favour of responsible release standards (with a focus on open source, but applicable to Gen AI as a whole): awo.cdn.ngo/media/document… Politico's reporting on Anthropic: politico.eu/article/eu-sup…

Anthropic postponing Mythos demonstrates how responsible release will be increasingly important as Gen AI gets more capable. Cybersecurity risks is just one Gen AI issue that can mitigated with staged release and external input into development (as we argue in a 2024 paper)

We hope that this can be useful in informing ongoing work on the protection of minors in the EU and UK, including the upcoming DFA and the AI Omnibus, and the enforcement of the DSA, AI Act and OSA. 8/8

Beyond that we identified gaps around issues such as child rights impact assessments, model repository moderation, the standardisation of AI-CSAM content reports, and child safety experts’ involvement in model development. 7/8

New paper alert 🚨: “Generative AI and Child Safety: What are the risks and how can we solve them?”, an NSPCC/AWO study, has just been published! It describes important gaps in EU and UK legislation relating to risks that children face from Gen AI. learning.nspcc.org.uk/media/1drdecrn… 1/8

Our Algorithm Governance Roundup is out! This month we spoke to Dave Buckley at @openminedorg, a non-profit building open-source PETs to audit proprietary systems. We spoke about several pilots with the Christchurch Call and UK @AISecurityInst.👇 eomail5.com/web-version?p=…

Had a wonderful chat with @finnmyrstad about tackling AI harms using consumer protection and DP law, for the Algo Gov newsletter last week! Check it out below😊

🙃ft.com/content/9350d0…

Provided some thoughts on the changes to Meta's Pay-or-Consent scheme to IAPP's @jduballreports! iapp.org/news/a/meta-re…

🚨NEW: The European Commission has just published the first draft of the Code of Practice for general-purpose #AI model providers. digital-strategy.ec.europa.eu/en/library/fir…

Great news! Today Commissioner-designate McGrath confirmed that the Digital Fairness Act will contain new rules to tackle addictive design features on social media, such as the endless scroll, autoplay and the endless streams of notifications — for everyone, not only minors.

An update to the context page covering the dangers of ad tech data on the open market and for sale to stalkers, bounty hunters and LEOs - context.center/topics/law-enf…

🚨 "Generative AI’s open source challenge: policy options to balance the risks and benefits of openness in AI regulation" by @NickBotton & @mathver is a MUST-READ for open-source enthusiasts. Key findings👇 1️⃣ "Lack of clarity regarding what constitutes “open source” in Generative AI has resulted in open washing; 2️⃣ Open washing disproportionately focuses on promoting the benefits of openness without fully addressing its risks; 3️⃣ Opening up access to external parties can improve risk mitigation measures; 4️⃣ An open science approach to releasing models can lead to increased safety; 5️⃣ Barriers exist that limit the potential of openness to external researchers; 6️⃣ Current policy approaches do not adequately tackle the openness challenge." ╰┈➤ There has been a positive buzz around open-source AI models, but we don't hear much about potential downsides and cases in which open-source might not be the best option. ╰┈➤ This paper does a great job of breaking down the risks and outlining a practical policy framework to help balance the risks & benefits associated with openness. Among policy options are: ➡️ "Threshold criteria for high-risk Generative AI models ➡️ Standards for responsible release ➡️ Systematic researcher vetting ➡️ A safe harbour for independent researchers ➡️ Subsidies for external research ➡️ Standards on levels of access ➡️ Due diligence requirements for model hosting platforms" ╰┈➤ I discuss open-source AI and its legal implications in my newsletter today; check it out below. 👉 Download the full paper below. 🏛️ STAY INFORMED: If you want to learn more about AI regulation & policy, I recommend you join 37,900+ people who subscribe to my AI governance newsletter, where I discuss the latest developments in the field, including outstanding research papers. If you have friends in this field, tell them to subscribe! (link below).

@jfelixcardoso Amazing! On my end I now roughly know the amount of Puerto Ricans in each purple state (that might have have been offended by the "garbage" comment). If it's any consolation, human memory has no know capacity limits, so one can never learn too much, really