As LLM-based agents increasingly browse the web on users' behalf, a natural question arises: can websites passively identify which underlying model powers an agent? Doing so would represent a significant security risk, enabling targeted attacks tailored to known model vulnerabilities. Across 14 frontier LLMs and four web environments spanning information retrieval and shopping tasks, we show that an agent's actions and interaction timings, captured via a passive JavaScript tracker, are sufficient to identify the underlying model with up to 96\% F1. We formalise this attack surface by demonstrating that classifiers trained on agent actions generalise across model sizes and families. We further show that strong classifiers can be trained from few interaction traces and that agent identity can be inferred early within an episode. Injecting randomised timing delays between actions substantially degrades classifier performance, but does not provide robust protection: a classifier retrained on delayed traces largely recovers performance. We release our harness and a labelled corpus of agent traces \href{https://github.com/KabakaWilliam/known_actions}{here}.

There are growing expectations that financial institutions and companies could, and should, play their part in finding solutions to biodiversity loss, for example, by investing in strategies that fully mitigate negative impacts on biodiversity and support the transition towards nature- positive economies. A key element of nature-related risk for investors and businesses is the "materiality" of their impacts on biodiversity. This includes financial, physical and reputational elements. Despite a proliferation of experimental tools for quantifying, assessing, and monitoring impacts on biodiversity (physical materiality), there is a gap as regards reputational materiality. Nature PRISM aims to complement existing approaches, by quantifying reputational materiality for industry, regulators, and investors. It is an automated data-driven tool, that assesses the perceived impact of corporate activity on biodiversity through a large-scale machine-learning analysis of publicly-available information. The tool is based on a Directed Acyclic Graph structured around a DAPSIR (Driver, Action, Pressure, State, Impact, Response) framework, providing a novel holistic approach to monitoring and reporting perceptions of biodiversity impacts. Importantly it picks up not just negative impacts, but also positive impacts and responses to biodiversity loss. We demonstrate the potential of Nature PRISM to reveal the extent of perceived correlations between drivers and biodiversity indicators across a wide range of sectors, using a novel dataset of over three million long-form news articles, identified through a set of expert- derived and industry-standard keywords. We also demonstrate the tool's potential to become a robust and comparable sector-specific indicator using a case study of housing developments in Europe. The tool has the potential to be a new and complementary evidence-based resource for investors, policymakers, and other stakeholders to assess reputational materiality and guide business decisions, disaggregatable by sector, geography, or individual company.

Unsustainable wildlife trade imperils thousands of species, but efforts to identify and reduce these threats are hampered by rapidly evolving commercial markets. Businesses trading wildlife-derived products innovate to remain competitive, and the patents they file to protect their innovations also provide an early-warning of market shifts. Here, we develop a novel machine-learning approach to analyse patent-filing trends and apply it to patents filed from 1970-2020 related to six traded taxa that vary in trade legality, threat level, and use type: rhinoceroses, pangolins, bears, sturgeon, horseshoe crabs, and caterpillar fungus. We found 27,308 patents, showing 130% per-year increases, compared to a background rate of 104%. Innovation led to diversification, including new fertilizer products using illegal-to-trade rhinoceros horn, and novel farming methods for pangolins. Stricter regulation did not generally correlate with reduced patenting. Patents reveal how wildlife-related businesses predict, adapt to, and create market shifts, providing data to underpin proactive wildlife-trade management approaches.

The surge in internet accessibility has transformed wildlife trade by facilitating the acquisition of wildlife through online platforms. This scenario presents unique ethical challenges for researchers, as traditional ethical frameworks for in-person research cannot be readily applied to the online realm. Currently, there is a lack of clearly defined guidelines for appropriate ethical procedures when conducting online wildlife trade (OWT) research. In response to this, we consulted the scientific literature on ethical considerations in online research and examined existing guidelines established by professional societies and ethical boards. Based on these documents, we present a set of recommendations that can inform the development of ethically responsible OWT research. Key ethical challenges in designing and executing OWT research include the violation of privacy rights, defining subjects and illegality, and the risk of misinterpretation or posing risks to participants when sharing data. Potential solutions include considering participants’ expectations of privacy, defining when participants are authors versus subjects, understanding the legal and cultural context, minimizing data collection, ensuring anonymization, and removing metadata. Best practices also involve being culturally sensitive when analyzing and reporting findings. Adhering to these guidelines can help mitigate potential pitfalls and provides valuable insights to editors, researchers, and ethical review boards, enabling them to conduct scientifically rigorous and ethically responsible OWT research to advance this growing field.

Wild species are under increasing threat, including from unsustainable harvest of wildlife products for commercial trade. Understanding how and why wildlife markets evolve is key to developing more proactive approaches to preventing unsustainable harvest. As in any other commercial sector, businesses who use wildlife-derived products must innovate to remain competitive, and patenting is a key process by which these innovations are protected from exploitation by others. Here we use machine learning methods to scrape and categorise the 27,308 patents filed between 1970 and 2020 that relate to new products or processes associated with six taxa variously threatened by wildlife trade, but with distinct differences in their uses, trade chains and geographies: rhinos, pangolins, bears, sturgeon, horseshoe crabs, and cordyceps caterpillar fungus. Our findings show increasing trends in patent filing for all taxa, above the background rate of increase for general global patent filing during our study period. Patents revealed diversifying product types, such as the emergence of agricultural products using rhinoceros horn; and processes to produce non-wild alternatives, such as methods for farming pangolins or synthesising sturgeon caviar. The introduction of wildlife trade regulations did not generally correlate with any decline in patenting rate, apart from a decrease in pangolin patenting following their inclusion in CITES Appendix I that halted international commercial trade in 2016. We show that patent data provide significant insights into how businesses view the future of the market for wildlife products. Regular, ongoing scans of patent data can reveal emerging trends in commercial interest in wild taxa, and identify novel product types and key entities who plan to commercialise different wildlife-derived products. Working with businesses to encourage innovation related to sustainable sourcing of, or alternatives to, products derived from threatened taxa presents a proactive approach towards harnessing commercial forces towards sustainability.

In this work, we investigate smart city technologies primarily through an examination of trends in patent filing. We apply machine learning methods both to explore the increasing rates of patent filing globally for smart city technologies, and also to identify the emerging topics on which companies are choosing to focus their efforts. We focus particularly on deployed and emerging urban systems-of-systems in China, which represent a high proportion of patents filed for smart city technologies, with a view to their potential global impacts. As a leading source of innovation in the development of smart cities, Chinese patent filing exerts significant influence on similar technologies adopted globally. Our global patent analysis highlights emerging trends in smart city innovations, and the increased adoption of technologies and processes that present significant human rights concerns, especially concerns to privacy, freedom of expression, and assembly.

This paper describes an ongoing experiment evaluating the efficacy of a digital safety intervention in six high-risk, low capacity Civil Society Organisations (CSOs) in Central Asia. The evaluation takes the form of statistical analysis of DNS traffic in each organisation, obtained via security tools installed by researchers. The hypothesis is that the digital safety intervention strengthens the overall digital security posture of the CSOs, as measured by number of malware attacks intercepted by a cloud-based DNS firewall installed on the CSOs networks. The research collects DNS traffic from CSOs that are participating in the digital safety intervention, and compares a treatment group consisting of four CSOs against DNS traffic from a second group of two CSOs in which the intervention has not yet taken place. This project is ongoing, with data collection underway at a number of Central Asian CSOs. In this paper we outline the experimental design of the project, and look at the early data coming out of the DNS firewall. This is done to support the ultimate question of whether DNS data such as this can be used to accurately assess the efficacy of digital hygiene efforts.

Key Points: 1. A rare example of a wildlife trade initiative that covers all stages of an evidence-based behaviour change intervention. 2. Intervention development involved combining extensive consumer research with human behaviour theory and past research. 3. Intervention used a cutting-edge, powerful combination of online news coverage and targeted advertising. 4. Post-intervention, 4% of the target audience changed their behaviour (vs 1% of non-target) and the intervention message was shown as the key cause; but high-level users did not decrease significantly pre-to post-intervention.

In this work we analyse the use of malicious mimicry and cloning of darknet marketplaces and other ‘onion services’ as means for phishing, akin to traditional ‘typosquatting’ on the web. This phenomenon occurs due to the complex trust relationships in Tor’s onion services, and particularly the complex webs of trust enabled by darknet markets and similar services. To do so, we built a modular scraper tool to identify networks of maliciously cloned darknet marketplaces; in addition to other characteristics of onion services, in aggregate. The networks of phishing sites identified by this scraper were then subject to clustering and analysis to identify the method of phishing and the networks of ownership across these sites. We present a novel discovery mechanism for sites, means for clustering and analysis of onion service phishing and clone sites, and an analysis of their spectrum of sophistication.

Conservationists increasingly use unstructured observational data, such as citizen science records or ranger patrol observations, to guide decision making. These datasets are often large and relatively cheap to collect, and they have enormous potential. However, the resulting data are generally "messy,'' and their use can incur considerable costs, some of which are hidden. We present an overview of the opportunities and limitations associated with messy data by explaining how the preferences, skills, and incentives of data collectors affect the quality of the information they contain and the investment required to unlock their potential. Drawing widely from across the sciences, we break down elements of the observation process in order to highlight likely sources of bias and error while emphasizing the importance of cross-disciplinary collaboration. We propose a framework for appraising messy data to guide those engaging with these types of dataset and make them work for conservation and broader sustainability applications.

Rapid changes in illicit drug demand, such as the Fentanyl epidemic, are a major public health issue. Policymakers currently rely on annual surveys to monitor public consumption, which are arguably too infrequent to detect rapid shifts in drug use. We present a novel method to predict drug use based on high-frequency sales data from darknet markets. We show that models based on historic trades alone cannot accurately predict drug demand. However, augmenting these models with data on Wikipedia page views for each drug greatly improves predictive accuracy, particularly for less popular drugs, suggesting such models may be particularly useful for detecting newly emerging substances. These results hold out-of-sample at high time frequency, across a range of drugs and countries. Therefore Wikipedia data may enable us to build a high-frequency measure of drug demand, which could help policymakers respond more quickly to future drug crises.

Human factors are increasingly recognised as central to conservation of biodiversity. Despite this, there are no existing systematic efforts to monitor global trends in perceptions of wildlife. With traditional news reporting now largely online, the internet presents a powerful means to monitor global attitudes towards species. In this work we develop a method using the Global Database of Events, Language, and Tone (GDELT) to scan global news media, allowing us to identify and download conservation-related articles. Applying supervised machine learning techniques, we filter irrelevant articles to create a continually updated global dataset of news coverage for seven target taxa: lion, tiger, saiga, rhinoceros, pangolins, elephants, and orchids, and observe that over two-thirds of articles matching a simple keyword search were irrelevant. We examine coverage of each taxa in different regions, and find that elephants, rhinos, tigers, and lions receive the most coverage, with daily peaks of around 200 articles. Mean sentiment was positive for all taxa, except saiga for which it was neutral. Coverage was broadly distributed, with articles from 73 countries across all continents. Elephants and tigers received coverage in the most countries overall, whilst orchids and saiga were mentioned in the smallest number of countries. We further find that sentiment towards charismatic megafauna is most positive in non-range countries, with the opposite being true for pangolins and orchids. Despite promising results, there remain substantial obstacles to achieving globally representative results. Disparities in internet access between low and high income countries and users is a major source of bias, with the need to focus on a diversity of data sources and languages, presenting sizable technical challenges...

Changing human behavior is essential for biodiversity conservation, but robust approaches for large scale change are needed. Concepts like repeat message exposure and social reinforcement, as well as mechanisms like online news coverage and targeted advertisements, are currently used by private and public sectors, and could prove powerful for conservation. Thus, to explore their potential in influencing wildlife consumption, we used online advertisements through Facebook, Google, and Outbrain, to promote news articles discussing the use of a Critically Endangered antelope (the Saiga tatarica) as a traditional Chinese medicine in Singapore. Our message, tailored to middle‐aged Chinese Singaporean women, framed saiga horn products as being no longer socially endorsed. Through advert performance and in‐depth analyses of Facebook user engagement, we assessed audience response. Our message pervaded Singapore's online media (e.g., our adverts were shown almost five million times; and the story ran on seven news outlets), and resulted in widespread desirable audience responses (e.g., 63% of Facebook users' engagements included identifiably positive features like calls for public action to reduce saiga horn consumption, anger at having unknowingly used a Critically Endangered species, and self‐pledges to no longer use it; only 13% of engagements included identifiably negative features). This work shows that targeted dissemination of online news articles can have promising results, and may have wide applicability to conservation.

Illegal wildlife trade is gaining prominence as a threat to biodiversity, but addressing it remains challenging. To help inform proactive policy responses in the face of uncertainty, in 2018 we conducted a horizon scan of significant emerging issues. We built upon existing iterative horizon scanning methods, using an open and global participatory approach to evaluate and rank issues from a diverse range of sources. Prioritized issues related to three themes: developments in biological, information, and financial technologies; changing trends in demand and information; and socioeconomic, geopolitical shifts and influences. The issues covered areas ranging from changing demographic and economic factors to innovations in technology and communications that affect illegal wildlife trade markets globally; the top three issues related to China, illustrating its vital role in tackling emerging threats. This analysis can support national governments, international bodies, researchers, and nongovernmental organizations as they develop strategies for addressing the illegal wildlife trade.

Since 2011, internet shutdowns have steadily become an increasingly popular form of digital repression, especially in India – which accounted for more than 50% of global recorded shutdowns from 2016 to 2019. Common shutdown justifications include ‘ensuring public safety’ in order to curb the prevalence of collective action in the form of protests and riots. This paper examines the correlation between internet shutdowns and a range of predictors, identifying riots as the main predictor of a shutdown. We focus on shutdowns throughout India between 2016 and 2019 with particular attention to Jammu and Kashmir. Primarily using data from the NGO Access Now and the Integrated Conflict Early Warning System, we apply Bayesian inference via generalised linear modelling implemented using the Stan probabilistic programming language, to estimate correlates of shutdown behaviour. We first examine how the prevalence of collective action may impact the probability of observing a shutdown; and second how the length of a shutdown impacts subsequent collective action. Our main finding is that riots seem to be the key predictor of a shutdown with increased protests and riots increasing the odds of observing a shutdown the same day by 7% with a 95% credible interval of 0.01-0.13 and 15% with a 95% credible interval of 0.03-0.26 respectively. As a predictor, however, the duration of an internet shutdown only has a marginal negative effect on the occurrence of riots at -8% per subsequent shutdown day with a 95% credible interval of -0.16 to -0.002.

An abstract is not available for this content. As you have access to this content, full HTML content is provided on this page. A PDF of this content is also available in through the ‘Save PDF’ action button.

The darknet is a network of websites that can be accessed only via special software that hides the details of the user’s connection, and also allows websites to be hosted without revealing their location or operator. Today, large-scale darknet marketplaces exist for illegal drugs, firearms, hacking tools, stolen identity documents, and a wide variety of other illicit goods. The darknet has not, to date, proven to be a particularly attractive platform for the buying and selling of illegal wildlife products. Despite this, the darknet provides a 'marketplace of last resort' that becomes increasingly attractive over other, more accessible, online services as law enforcement and platform operators enforce policies against trading in illegal wildlife products. This makes the ongoing study of darknet markets an important avenue for research as other policies against online illegal wildlife trading emerge.

Anecdotal evidence suggests an increasing number of people are turning to VPN services for the properties of privacy, anonymity and free communication over the internet. Despite this, there is little research into what these services are actually being used for. We use DNS cache snooping to determine what domains people are accessing through VPNs. This technique is used to discover whether certain queries have been made against a particular DNS server. Some VPNs operate their own DNS servers, ensuring that any cached queries were made by users of the VPN. We explore 3 methods of DNS cache snooping and briefly discuss their strengths and limitations. Using the most reliable of the methods, we perform a DNS cache snooping scan against the DNS servers of several major VPN providers. With this we discover which domains are actually accessed through VPNs. We run this technique against popular domains, as well as those known to be censored in certain countries; China, Indonesia, Iran, and Turkey. Our work gives a glimpse into what users use VPNs for, and provides a technique for discovering the frequency with which domain records are accessed on a DNS server.

In March 2018, to the surprise of many users, the largest Reddit forums related to darknet markets (DNM) were banned overnight. For users, whose trading activity relied heavily on these forums, the ban was a threat to the community as a whole. In this study we use a complete set of posts from the newly founded forums in the darknet-based “Dread” platform to examine key discussion topics and the sentiment of the community towards the ban. We look at the level of user engagement in the new forums, and the number of users who retain their old usernames. Applying topic modelling to posts on the new forum, we show that there are many overlapping themes across both the banned and new forums, and that discussions on drugs are the most prominent, followed by vendors, shipping reviews, and payment methods. We observe that the new community demonstrates negative sentiment toward the unexpected ban and the loss of accumulated information, but also holds a favourable view of Dread in the hopes that it will offer greater features and security for the users. Users across both platforms express attachment and affinity to the general DNM community, and demonstrate relation to it beyond the commercial purpose.

The growing public nature of academic journals along with current best practices of sharing primary data for scientific research are profoundly valuable for the understanding of a species and their conservation efforts. On the other hand, public spatial data on endangered species may be easily abused by wildlife criminals. In this paper, we discuss how geo-indistinguishability, a formal notion of privacy for location-based systems, can be used to add noise to published spatial data whilst allowing quantification of such tradeoff.