iiinova — where your digital self keeps evolving

Mia Hoffmann @miahoffmann.bsky.social · 11/09/2025

🤖✨ New report with @partnershipai.bsky.social! AI agents pose new risks. Monitoring is essential to ensure effective oversight and intervention when needed. Our paper presents a framework for real-time failure detection that takes into account stakes, reversibility and affordances of agent actions.

111

Reposted by Mia Hoffmann

CSET @csetgeorgetown.bsky.social · 30/07/2025

✨New Analysis✨ Can the new EU AI Code of Practice change the global AI safety landscape? As companies like Anthropic, OpenAI, and Google sign on, CSET’s @miahoffmann.bsky.social explores the code’s Safety and Security chapter. cset.georgetown.edu/article/eu-a...

cset.georgetown.edu

AI Safety under the EU AI Code of Practice — A New Global Standard? | Center for Security and Emerging Technology

To protect Europeans from the risks posed by artificial intelligence, the EU passed its AI Act last year. This month, the EU released a Code of Practice to help providers of general purpose AI comply ...

012

Reposted by Mia Hoffmann

Vikram Venkatram @vikramvenkatram.bsky.social · 24/07/2025

Yesterday's new AI Action Plan has a lot worth discussing! One interesting aspect is its statement that the federal government should withhold AI-related funding from states with "burdensome AI regulations." This could be cause for concern.

163

Reposted by Mia Hoffmann

CSET @csetgeorgetown.bsky.social · 28/05/2025

⚖️ New Explainer! Effectively evaluating AI models is more crucial than ever. But how do AI evaluations actually work? In their new explainer, @jessicaji.bsky.social, @vikramvenkatram.bsky.social & @stephbatalis.bsky.social break down the different fundamental types of AI safety evaluations.

142

Reposted by Mia Hoffmann

Helen Toner @hlntnr.bsky.social · 19/05/2025

💡Funding opportunity—share with your AI research networks💡 Internal deployments of frontier AI models are an underexplored source of risk. My program at @csetgeorgetown.bsky.social just opened a call for research ideas—EOIs due Jun 30. Full details ➡️ cset.georgetown.edu/wp-content/u... Summary ⬇️

195

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

10) If you’re still doubting the benefits of AI incident tracking, come by the Massive Data Institute’s event on "AI Hazards: Understanding AI Incidents" today at 3pm, and let me and my fabulous co-panelists convince you in person! mdi.georgetown.edu/events/tswee...

mdi.georgetown.edu

Tech & Society Week 2025 — AI Hazards: Understanding AI Incidents - Massive Data Institute

On Monday, March 17, 2025 from 3:00 to 4:00pm in Fisher Colloquium in Hariri Building on the Hilltop Campus, we will be hosting a panel discussion on AI incidents during Tech & Society Week 2025. The ...

100

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

Finally, and critically: central data collection and dissemination of lessons learned means that harms only have to occur once for everyone to mitigate their risk. This prevents recurrence and builds user and consumer confidence, which is essential for widespread AI adoption.

100

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

Incident tracking also reveals new, unexpected AI failure modes that we aren’t yet mitigating against. Over time, systematic data collection can help detect emerging risks and new types of harms, a critical benefit given the fast pace of AI innovation and deployment.

100

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

Over time, incident data can be used to evaluate the effectiveness of new safety policies and regulation through before and after comparisons. This helps refine governance policies through a direct feedback loop.

100

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

Using real-world data on what works and what doesn’t to guide AI safety research will help us innovate quicker and build reliable systems that are safe to deploy faster. In this way, incident reporting can help prioritize and direct AI safety research to where it is most effective.

100

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

AI incidents also shed light on the effectiveness of existing safety efforts. We might learn where current technical standards or risk management processes are insufficient to protect people from harm, revealing critical gaps that can be addressed by AI safety research.

110

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

For instance, we can learn about *how* the use of AI results in harm, e.g. through misuse, user error or AI failure. This information helps channel resources to the right kinds of safety efforts, since preventing misuse requires different measures than addressing operator error.

110

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

Why should the government do this? What makes AI risk management so tricky is predicting how deploying an AI system can go wrong. AI incidents are a rich source of information about AI harms, harm mechanisms, AI failure modes and more. Leveraging those insights can make AI use safer.

110

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

Broadly speaking, an AI incident reporting regime has 4 core parts: 1) Incident detection; 2) Reporting to oversight bodies and inclusion in incident database; 3) Performance of impact assessments and root cause analyses; and 4) Dissemination of lessons learned

110

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

First, a definition. AI incidents are situations in which a deployed AI system is implicated in harm, e.g. when an AI recruiting tool makes a biased hiring decision. Incidents are varied and often take unexpected forms, so go check out the AIID for more real-world examples! incidentdatabase.ai

incidentdatabase.ai

Welcome to the Artificial Intelligence Incident Database

The starting point for information about the AI Incident Database

110

Mia Hoffmann @miahoffmann.bsky.social · 17/03/2025

Today, @csetgeorgetown.bsky.social published our recommendations for the U.S. AI Action Plan. One of them is a CSET evergreen: implement an AI incident reporting regime for AI used by the federal government. Why? Short answer: because we can learn a ton from incidents! Long answer: 👇

142

Reposted by Mia Hoffmann

CSET @csetgeorgetown.bsky.social · 14/03/2025

🚨We're hiring — only a few days left to apply!🚨 CSET is looking for a Media Engagement Specialist to amplify our research. If you're a strategic communicator who can craft press releases, media pitches, & social content, apply by March 17, 2025! cset.georgetown.edu/job/media-en...

cset.georgetown.edu

Media Engagement Specialist | Center for Security and Emerging Technology

The Center for Security and Emerging Technology, under the School of Foreign Service, is a research organization focused on studying the security impacts of emerging technologies, supporting academic ...

002

Reposted by Mia Hoffmann

CSET @csetgeorgetown.bsky.social · 12/03/2025

What: CSET Webinar 📺 When: Tuesday, 3/25 at 12PM ET 📅 What’s next for AI red-teaming? And how do we make it more useful? Join Tori Westerhoff, Christina Liaghati, Marius Hobbhahn, and CSET's @dr-bly.bsky.social * @jessicaji.bsky.social for a great discussion: cset.georgetown.edu/event/whats-...

cset.georgetown.edu

What’s Next for AI Red-Teaming? | Center for Security and Emerging Technology

On March 25, CSET will host an in-depth discussion about AI red-teaming — what it is, how it works in practice, and how to make it more useful in the future.

044

Reposted by Mia Hoffmann

CSET @csetgeorgetown.bsky.social · 10/03/2025

What does the EU's shifting strategy mean for AI? CSET's @miahoffmann.bsky.social & @ojdaniels.bsky.social have a new piece out for @techpolicypress.bsky.social. Read it now 👇

044

Reposted by Mia Hoffmann

Tech Policy Press @techpolicypress.bsky.social · 10/03/2025

Mia Hoffmann and Owen J. Daniels from Georgetown’s Center for Security and Emerging Technology say Europe's apparent shift on AI policy could change the global landscape for AI governance.

buff.ly

Out of Balance: What the EU's Strategy Shift Means for the AI Ecosystem | TechPolicy.Press

Mia Hoffmann and Owen J. Daniels from Georgetown’s Center for Security and Emerging Technology say Europe's movements could change the global landscape.

063

Mia Hoffmann @miahoffmann.bsky.social · 10/03/2025

If you’ve ever wondered what the EU and elephants have in common - or are wondering now- read my latest piece with @ojdaniels.bsky.social! We take a look what the EU’s new innovation-friendly regulatory approach might mean for the global AI policy ecosystem www.techpolicy.press/out-of-balan...

techpolicy.press

Out of Balance: What the EU's Strategy Shift Means for the AI Ecosystem | TechPolicy.Press

Mia Hoffmann and Owen J. Daniels from Georgetown’s Center for Security and Emerging Technology say Europe's movements could change the global landscape.

021

Reposted by Mia Hoffmann

CSET @csetgeorgetown.bsky.social · 03/03/2025

CSET is hiring 📢 We’re hiring a software engineer to support @emergingtechobs.bsky.social. Help build high-quality public tools and datasets to inform critical decisions on emerging tech issues. Interested or know someone who would be? Learn more and apply 👇 cset.georgetown.edu/job/software...

cset.georgetown.edu

Software Engineer | Center for Security and Emerging Technology

The Center for Security and Emerging Technology (CSET), under the School of Foreign Service, is hiring a Software Engineer. The Software Engineer will be a generalist who can flex between full-stack w...

031

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

Thirdly, and most importantly, this decision reveals that the new European Commission is buying into the false narrative of innovation versus regulation which already dominates - and paralyzes - US tech policy.

000

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

Secondly, these questions will now be relegated to the national legal systems, which means uneven rules across the EU. Because what opponents to EU regulation need to understand is that the alternative to EU rules is not No Rules, it is 27 different sets of rules. How’s that for simplification?

100

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

So what does this mean? First, substantively, the questions on how to deal with liability for fundamental rights violations from AI, and liability across the value chains will remain open at the EU level.

100

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

For example, the PLD covers material harms from AI, but violations of fundamental rights - which are covered in the EU AI Act - would have fallen in the domain of the AILD. Similarly, the AILD was going to address the question of how liability should be distributed along the AI value chain.

100

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

Now, the AILD was not a flawless proposal. There was a lot of overlap with the Product Liability Directive (PLD), which already deals with software, including AI. But at the same time, it dealt with important aspects the PLD did not.

100

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

It appears the decision was political, and is a reflection of the new orientation of the new Commission: industry-friendly and anti-regulation. This is mirrored in statements made by EU leaders at the AI summit, claiming that EU rules would be “simplified” and applied in “business-friendly ways”.

100

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

The Commission cited “no foreseeable agreement” on the proposal as the reason for dropping it. But the decision follows months of lobbying against the directive from industry, e.g. just in January the American Chamber of Commerce released a position paper calling for the withdrawal of the AILD.

100

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

The AILD was proposed in 2022 and its purpose was to "improve the functioning of the internal market by laying down uniform rules for certain aspects of non-contractual civil liability for damage caused with the involvement of AI systems.“ I.e. create EU-wide liability rules for harms from AI.

100

Mia Hoffmann @miahoffmann.bsky.social · 13/02/2025

There have been a ton of AI policy developments coming out of the EU these past weeks, but one deeply concerning one is the withdrawal of the AI Liability Directive (AILD) by the European Commission. Here’s why:

133

Mia Hoffmann @miahoffmann.bsky.social · 03/02/2025

The next round of provisions comes into force in 6 months, on 2 August 2025. This is when the regulation’s penalties and the rules for general-purpose AI become applicable. Also by then, the oversight and governance structures at EU and member states level must be established.

000

Mia Hoffmann @miahoffmann.bsky.social · 03/02/2025

(8) The use of “real-time” remote biometric identification systems in public spaces (with law enforcement exceptions). Once a year, the European Commission is tasked with assessing whether the list needs to be amended, so the use cases are potentially subject to change.

100

Mia Hoffmann @miahoffmann.bsky.social · 03/02/2025

(4) AI systems for predictive policing (5) AI-powered web- or CCTV scraping to build facial recognition databases (6) Emotion recognition AI in workplaces or schools (7) Biometric categorization to infer personal characteristics (with law enforcement exceptions).

100

Mia Hoffmann @miahoffmann.bsky.social · 03/02/2025

Secondly, the following AI uses are believed to pose unacceptable risks to health, safety and fundamental rights, and are from now on banned in the EU: (1) Deceptive or manipulative AI systems (2) AI systems exploiting vulnerabilities (3) AI systems for “social scoring”

100

Mia Hoffmann @miahoffmann.bsky.social · 03/02/2025

AI literacy is defined as the skills, knowledge and understanding that allow developers, deployers and affected persons to make an informed deployment of AI systems, as well as to gain awareness about the opportunities and risks of AI and possible harm it can cause.

100

Mia Hoffmann @miahoffmann.bsky.social · 03/02/2025

From now on, AI developers and deployers are required to ensure a sufficient level of AI literacy of their staff (or anyone operating the AI system on their behalf), taking into account their expertise and training, the context of use and the people affected by the AI system.

100

Mia Hoffmann @miahoffmann.bsky.social · 03/02/2025

First, the General Provisions. They cover the purpose, scope and definitions of the AI Act, but - and this is important - they also include an AI literacy requirement.

100

Mia Hoffmann @miahoffmann.bsky.social · 03/02/2025

Yesterday, the EU AI Act’s first few provisions came into effect. The General Provisions and the prohibitions of unacceptable risk AI systems are applicable from now on. Here’s what that means:

100

Mia Hoffmann @miahoffmann.bsky.social · 29/01/2025

It should also consider how widely US models are being adopted everywhere in the world, since diffusion is an opportunity to set informal norms and standards on AI, especially in contexts where our norms might diverge from others’ (e.g. democratic values, freedom of speech).

000

Mia Hoffmann @miahoffmann.bsky.social · 29/01/2025

Outside of the US, in particular, adopters don’t have a strong incentive to “buy American”, especially not over a free/much cheaper Chinese alternative. Our definition of AI leadership should therefore encompass much more than just being in the lead on “AGI”.

100

Mia Hoffmann @miahoffmann.bsky.social · 29/01/2025

This means that even if a private American lab’s AI model maintains benchmark leadership, the most adopted and widely used models are going to be free, open (source) and potentially non-US models, if their performance is only marginally worse.

100

Mia Hoffmann @miahoffmann.bsky.social · 29/01/2025

Businesses and organizations adopting AI will also care about other aspects, such as the convenience of flexibly adapting a base model to their needs, or the advantage of not having to send their input data to a hosted model for processing.

100

Mia Hoffmann @miahoffmann.bsky.social · 29/01/2025

Businesses and organizations interested in adopting an AI product will care way more about cost than slight performance differences on a benchmark. And what DeepSeek R1 has shown is that through innovative design comparable performance can be achieved at much lower cost.

100

Mia Hoffmann @miahoffmann.bsky.social · 29/01/2025

Instead of focusing on benchmark performance of individual models, where differences these days have become marginal and rankings change within months if not weeks, we should shift our attention towards AI diffusion

100