In this talk I will focus on analysing the robustness of systems which depend on crowdsourced and sensor data. I will showcase vulnerabilities on mobile crowdsourcing services which can be exploited to launch data poisoning attacks successful in faking online posts for robberies, gunshots, and other dangerous incidents, faking fitness activities with supernatural speeds and distances. I will then show how data poisoning can impact 3D object detection in sensor-rich autonomous vehicles and discuss strategies for detecting such issues.

https://cam-ac-uk.zoom.us/j/81982791343?pwd=MytlVUh6c01OTGpWeVJKNGlKZklzZz09

Meeting ID: 819 8279 1343
Passcode: 079963

RECORDING : Please note, this event will be recorded and will be available after the event for an indeterminate period under a CC BY -NC-ND license. Audience members should bear this in mind before joining the webinar or asking questions.

NOTE: Please do not post URLs for the talk, and especially Zoom links to Twitter because automated systems will pick them up and disrupt our meeting.

This talk introduces Mysticeti a byzantine consensus protocol with low-latency and high resource efficiency. It leverages a DAG based on Threshold Clocks and incorporates innovations in pipelining and multiple leaders to reduce latency in the steady state and under crash failures. Mysticeti is the first byzantine protocol to achieve WAN latency of 0.5s for consensus commit, at a throughput of over 50k TPS that matches the state-of-the-art. Additionally, and if time permits, this talk describes a variant of Mysticeti, called Mysticeti-FPC, that incorporates a fast commit path that has even lower latency by forgoing consensus whenever possible.

https://cam-ac-uk.zoom.us/j/86368615412?pwd=V1Ayb2JiR0UxOCtnbDlIQ0dIc3ZNQT09
Meeting ID: 863 6861 5412
Passcode: 160590

RECORDING: Please note, this event will be recorded and will be available after the event for an indeterminate period under a CC BY -NC-ND license. Audience members should bear this in mind before joining the webinar or asking questions.

Note: Please do not post URLs for the talk, and especially Zoom links to Twitter because automated systems will pick them up and disrupt our meeting.

Large language models (LLMs) can "lie", which we define as outputting false statements despite "knowing" the truth in a demonstrable sense. LLMs might "lie", for example, when instructed to output misinformation. Here, we develop a simple lie detector that requires neither access to the LLM's activations (black-box) nor ground-truth knowledge of the fact in question. The detector works by asking a predefined set of unrelated follow-up questions after a suspected lie, and feeding the LLM's yes/no answers into a logistic regression classifier. Despite its simplicity, this lie detector is highly accurate and surprisingly general. When trained on examples from a single setting -- prompting GPT-3.5 to lie about factual questions -- the detector generalises out-of-distribution to (1) other LLM architectures, (2) LLMs fine-tuned to lie, (3) sycophantic lies, and (4) lies emerging in real-life scenarios such as sales. These results indicate that LLMs have distinctive lie-related behavioural patterns, consistent across architectures and contexts, which could enable general-purpose lie detection.

https://cam-ac-uk.zoom.us/j/88053652228?pwd=NG1LTDdUc2VkV3pGdlpSdHZ5N3h0Zz09

Meeting ID: 880 5365 2228
Passcode: 081966

RECORDING : Please note, this event will be recorded and will be available after the event for an indeterminate period under a CC BY -NC-ND license. Audience members should bear this in mind before joining the webinar or asking questions.

NOTE : Please do not post URLs for the talk, and especially Zoom links to Twitter because automated systems will pick them up and disrupt our meeting.

The talk presents open problems in the study of machine unlearning. The need for machine unlearning, i.e., obtaining a model one would get without training on a subset of data, arises from privacy legislation and as a potential solution to data poisoning or copyright claims. The first part of the talk discusses approaches that provide exact unlearning: these approaches output the same distribution of models as would have been obtained by training without the subset of data to be unlearned in the first place. While such approaches can be computationally expensive, we discuss why it is difficult to relax the guarantee they provide to pave the way for more efficient approaches. The second part of the talk asks if we can verify unlearning. Here we show how an entity can claim plausible deniability when challenged about an unlearning request that was claimed to be processed, and conclude that at the level of model weights, being unlearnt is not always a well-defined property. Instead, unlearning is an algorithmic property.

RECORDING : Please note, this event will be recorded and will be available after the event for an indeterminate period under a CC BY -NC-ND license. Audience members should bear this in mind before joining the webinar or asking questions.

https://cam-ac-uk.zoom.us/j/82112795708?pwd=VFBTVjI1YkRqMXY5MEpRcXYzdmN6QT09

Meeting ID: 821 1279 5708
Passcode: 468381

NOTE: Please do not post URLs for the talk, and especially Zoom links to Twitter because automated systems will pick them up and disrupt our meeting.

In this talk, I will first review three decades of research in the field of password-authenticated key exchange (PAKE). PAKE protocols can be categorized into two types: balanced and augmented schemes. I will share my experience of designing a balanced PAKE called J-PAKE in 2008 (joint work with Ryan). Today, J-PAKE has been deployed in many real-world applications, e.g., Google Nest, ARM Mbed, Amazon Fire stick and Thread products.

Next, I will focus on augmented PAKE, which is a different challenge. Today, SRP-6a is the only augmented PAKE that has enjoyed wide use, e.g., in Apple's iCloud, 1Password and Proton mail. Limitations of SRP-6a, such as heuristic security, a lack of efficiency (due to the mandated use of a safe prime) and a lack of support for elliptic curve implementations are well-known, but for the past 25 years, there seems to be no better alternative. In 2020, IETF chose OPAQUE as an augmented PAKE standard, but open issues leave it unclear whether OPAQUE will replace SRP-6a.

Finally, I will present Owl, a new augmented PAKE (joint work with Bag, Chen and van Oorshot; see https://eprint.iacr.org/2023/768). Owl is obtained by efficiently adapting J-PAKE to an augmented setting. While J-PAKE is symmetric, Owl is asymmetric. Both protocols follow the same design principle but they are suitable for different applications. I will show that Owl is systematically better than SRP-6a in every aspect, including security computation, communication, message sizes and cryptographic agility. Owl is also free from several security and implementation issues faced by OPAQUE.

https://cam-ac-uk.zoom.us/j/88950422934?pwd=WHJsSklROW90YVVxbndQYTlJTERIUT09

Meeting ID: 889 5042 2934
Passcode: 853480

RECORDING : Please note, this event will be recorded and will be available after the event for an indeterminate period under a CC BY -NC-ND license. Audience members should bear this in mind before joining the webinar or asking questions.

NOTE: Please do not post URLs for the talk, and especially Zoom links to Twitter because automated systems will pick them up and disrupt our meeting.

The public key infrastructure that secures the web has been around for nearly three decades. Since 2012, it has become a critical (albeit unappreciated) aspect of daily life for billions of people. In that short time, a dizzying number of technologies to improve security and privacy on the web have been designed, deployed, and, in many cases, deprecated. We’ll look at those which have become fundamental to online security, those which didn’t work out in practice, and the unsolved research problems remaining. We’ll also peek behind the curtain to see how contemporary realpolitik between countries over their ‘digital sovereignty’, profit incentives of corporate stakeholders and increasingly expansive government regulations threaten the WebPKI as it exists today.

Note: Please do not post URLs for the talk, and especially Zoom links to Twitter because automated systems will pick them up and disrupt our meeting.

https://cam-ac-uk.zoom.us/j/81403861657?pwd=aWo3Q0pFSDgwbWwxdHRtNVcyNkkrUT09

Meeting ID: 814 0386 1657
Passcode: 578528

Software constantly evolves in response to user needs: new features are built, deployed, mature and grow old, and eventually their usage drops enough to merit switching them off. In any large codebase, this feature lifecycle can naturally lead to retaining unnecessary code and data. Removing these respects users’ privacy expectations, as well as helping engineers to work efficiently. In prior software engineering research, we have found little evidence of code deprecation or dead-code removal at industrial scale. We describe Systematic Code and Asset Removal Framework (SCARF), a product deprecation system to assist engineers working in large codebases. SCARF identifies unused code and data assets and safely removes them. It operates fully automatically, including committing code and dropping database tables. It also gathers developer input where it cannot take automated actions, leading to further removals. Dead code removal increases the quality and consistency of large codebases, aids with knowledge management and improves reliability. SCARF has had an important impact at Meta. In the last year alone, it has removed petabytes of data across 12.8 million distinct assets, and deleted over 104 million lines of code.

https://cam-ac-uk.zoom.us/j/87445737656?pwd=RCs2aWplOTBudzV6SGlPa3Bub0ZJZz09

Meeting ID: 874 4573 7656
Passcode: 290144

RECORDING : Please note, this event will be recorded and will be available after the event for an indeterminate period under a CC BY -NC-ND license. Audience members should bear this in mind before joining the webinar or asking questions.

Note: Please do not post URLs for the talk, and especially Zoom links to Twitter because automated systems will pick them up and disrupt our meeting.

European lawmakers have ruled through the Digital Markets Act that users on different platforms should be able to exchange messages with each other. Yet messaging interoperability opens up a Pandora’s box of security and privacy challenges. While championed not just as an antitrust measure but as a means of providing a better experience for the end user, interoperability runs the risk of making the user experience worse if poorly executed. There are two fundamental questions: how to enable the actual message exchange, and how to handle the numerous residual challenges arising from encrypted messages passing from one service provider to another—including but certainly not limited to content moderation, user authentication, key management, and metadata sharing between providers.

In this talk, we will survey specific open questions and challenges in interoperable end-to-end encrypted messaging, with a particular focus on key management, user identity, and content moderation. We will outline existing protocols and designs, discuss where current solutions fall short, and explore possible ways of tackling these challenges.

https://cam-ac-uk.zoom.us/j/83553682797?pwd=Nnh1cklrRTVoaTJ2Y2tMVDFJS09qdz09

Meeting ID: 835 5368 2797
Passcode: 811133

NOTE: Please do not post URLs for the talk, and especially Zoom links to Twitter because automated systems will pick them up and disrupt our meeting.

Conspiracy theories have become a pervasive and potent force in the digital age, challenging societies and democracies worldwide.
This talk delves into the enigmatic origins of the QAnon conspiracy to offer a comprehensive analysis of the online fringe communities that facilitate such discourse.
We employ a data-driven cross-platform mixed-methods approach to investigate the evolution, behavior, and impact of QAnon across various alternative social networks.

Conspiracy theories, including QAnon, pose significant threats to democracies and individual autonomy.
This is exemplified by their exploitation for political gain, as evidenced by events such as the 2016 US Presidential Elections and the 2021 insurrection at the US Capitol.
Furthermore, these theories have real-world consequences, from public health threats due to COVID-19 misinformation to incidents of violence and radicalization.

Our research transcends QAnon's specific narratives to address critical questions about the movement, including but not limited to the activity of adherents, discussion topics, and community responses to platform shutdowns and online migration.
This talk underscores the imperative of understanding conspiracy theories in a digital world and the urgent need to develop strategies for countering their influence.
We provide unique insights into the dynamics of online communities, the challenges of moderation, and the intricate interplay between conspiracy theories and alternative social networks.

https://us02web.zoom.us/j/86421076573?pwd=aG4rY2VyZmttcVBhOTgvd2lXZ09NZz09
Meeting ID: 864 2107 6573
Passcode: 825423

RECORDING : Please note, this event will be recorded and will be available after the event for an indeterminate period under a CC BY -NC-ND license. Audience members should bear this in mind before joining the webinar or asking questions.

2-5% of global GDP – some $2-5T USD – is estimated to be associated with economic crime. Less than 1% of this is seized, while the total cost of compliance is around $274B USD: we are not winning this fight. This talk describes the work we do at Mastercard that seeks to move past the legacy approaches to economic crime, both fraud and money laundering, and questions the dominant theories of change in this space.