AI’s right to forget – Machine unlearning

Machine unlearning is a growing field within AI that aims to address the challenge of forgetting outdated, incorrect, or private data in machine learning (ML) models. ML models struggle to forget information, which has significant implications for privacy, security, and ethics. This has led to the development of machine unlearning techniques.

When issues arise with a dataset, it is possible to modify or delete the dataset. However, if the data has been used to train an ML model, it becomes difficult to remove the impact of a problematic dataset. ML models are often considered black boxes, making it challenging to understand how specific datasets influenced the model and undo their effects.

OpenAI has faced criticism for the data used to train their models, and generative AI art tools are involved in legal battles regarding their training data. This highlights concerns about privacy and the potential disclosure of information about individuals whose data was used to train the models.

Machine unlearning aims to erase the influence of specific datasets on ML systems. This involves identifying problematic datasets and excluding them from the model or retraining the entire model from scratch. However, the latter approach is costly and time-consuming.

Efficient machine unlearning algorithms are needed to remove datasets without compromising utility. Some promising approaches include incremental updates to ML systems, limiting the influence of data points, and scrubbing network weights to remove information about specific training data.

However, machine unlearning faces challenges, including efficiency, standardization of evaluation metrics, validation of efficacy, privacy preservation, compatibility with existing ML models, and scalability to handle large datasets.

To address these challenges, interdisciplinary collaboration between AI experts, data privacy lawyers, and ethicists is required. Google has launched a machine unlearning challenge to unify evaluation metrics and foster innovative solutions.

Looking ahead, advancements in hardware and infrastructure will support the computational demands of machine unlearning. Collaborative efforts between legal professionals, ethicists, and AI researchers can align unlearning algorithms with ethical and legal standards. Increased public awareness and potential policy and regulatory changes will also shape the development and application of machine unlearning.

Businesses using large datasets are advised to understand and adopt machine unlearning strategies to proactively manage data privacy concerns. This includes monitoring research, implementing data handling rules, considering interdisciplinary teams, and preparing for retraining costs.

Machine unlearning is crucial for responsible AI, improving data handling capabilities while maintaining model quality. Although challenges remain, progress is being made in developing efficient unlearning algorithms. Businesses should embrace machine unlearning to manage data privacy issues responsibly and stay up-to-date with advancements in the field.

Read more

Employees at Fortune 1000 telecom companies are some of the most exposed on darkweb, researchers report

A recent report by threat intelligence firm SpyCloud has shed light on the alarming vulnerability of employees at Fortune 1000 telecommunications companies on dark web sites. The report reveals that researchers have uncovered approximately 6.34 million pairs of credentials, including corporate email addresses and passwords, which are likely associated with employees in the telecommunications sector.

The report highlights this as an ‘extreme’ rate of exposure compared to other sectors. In comparison, SpyCloud’s findings uncovered 7.52 million pairs of credentials belonging to employees in the tech sector, but this encompassed a significantly larger pool of 167 Fortune 1000 companies.

Media reports that these findings underscore the heightened risk faced by employees within the telecommunications industry, as their credentials are more readily available on dark web platforms. The compromised credentials pose a significant threat to the affected individuals and their respective companies, as cybercriminals can exploit them for various malicious activities such as unauthorized access, data breaches, and targeted attacks.

Western Digital, a technology company, confirms that hackers stole customer data

Western Digital, a technology company, has notified its customers after the March 2023 data breach and confirmed that the customer data was stolen.

In a press release, the company mentioned it worked with external forensic experts and determined that the hackers obtained a copy of a database which contained limited personal information of online store customers. The exact number of affected customers has not been disclosed. The company has notified affected customers and advised them to remain vigilant against potential phishing attempts.

The March data breach had previously been reported in early April when the company disclosed it has suffered a cyberattack. TechCrunch reported that an ‘unnamed’ hacking group breached Western Digital, claiming to have stolen ten terabytes of data.

The hackers subsequently published some of the stolen data and threatened to release more if their demands were not met. Western Digital has restored the majority of its impacted systems and services and continues to investigate the incident.

Ransomware criminal group leaks MSI’s private code on darkweb

The ransomware gang responsible for targeting Taiwanese PC manufacturer MSI has leaked the private code signing keys of the company available on their darkweb leak site. The attack, orchestrated by the group known as Money Message, was announced in early April: The group revealed that they had successfully breached the systems of MSI, a multinational IT corporation renowned for its production and distribution of motherboards and graphics cards worldwide, including in the USA and Canada. MSI is headquartered in Taipei, Taiwan.

It is reported that initially, the criminal group demanded a ransom from MSI, threatening to publish the stolen files if their demands were not met by a specified deadline. However, the group has eventually exposed MSI’s private code signing keys on their darkweb leak site. These keys are of significant importance as they are used to authenticate the legitimacy and integrity of software and firmware updates released by the company. Malicious actors could potentially misuse these keys to distribute malware or carry out other malicious activities, putting MSI’s customers at risk. The company now faces the daunting task of mitigating the potential fallout from this exposure and bolstering their cybersecurity measures to prevent further unauthorized access.

ICANN launches project to look at what drives malicious domain name registrations

The Internet Corporation for Assigned Names and Numbers (ICANN) has launched a project to explore the practices and choices of malicious actors when they decide to use the domain names of certain registrars over others. The project, called Inferential Analysis of Maliciously Registered Domains (INFERMAL), will systematically analyse the preferences of cyberattackers and possible measures to mitigate malicious activities across top-level domains (TLDs). It is funded as part of ICANN’s Domain Name System (DNS) Security Threat Mitigation Program, which aims to reduce the prevalence of DNS security threats across the Internet.

The team leading the project intends to collect and analyse a comprehensive list of domain name registration policies pertinent to would-be attackers, and then use statistical modelling to identify the registration factors preferred by attackers. It is expected that the findings of the project could help registrars and registries identify relevant DNS anti-abuse practices, strengthen the self-regulation of the overall domain name industry, and reduce the costs associated with domain regulations. The project would also help increase the security levels of domain names and, thus, the trust of end-users.

Data poisoning – a new type of cyberattacks against AI systems

Data poisoning is a new type of cyber-attack aimed at misleading AI systems. AI is developed by processing huge amounts of data. The quality of data impacts the quality of AI. Data poisoning is the intentional supply of wrong or misleading data to impact the quality of AI. Data poisoning is becoming particularly risky with the development of Large Language Models (LLM) such as ChatGPT.

Researchers from the Swiss Federal Institute of Technology (ETH) in Zurich, Google, NVIDIA and Robust Intelligence have recently published a preprint paper investigating the feasibility of data poisoning attacks against machine learning (ML) models used in artificial intelligence (AI). They injected corrupted data into an existing training data set in order to influence the behaviour of an AI algorithm that is being trained on it. It impacted the functionality of AI systems.

As AI systems are becoming more complex and massive, the detection of data poisoning attacks will be difficult. The main risks are in dealing with politically charged topics.

First issue of Commonwealth Cybercrime Journal highlights AI use in judicial decision-making, among other topics

The Commonwealth has recently released its first issue of a cybercrime journal to draw attention to policy-influencing articles and commentary by academics, policymakers, practitioners, and experts exploring significant cybercrime and cybersecurity issues. This first issue underscores regional cybercrime trends in Africa, the Caribbean, Southeast Asia and the UK, highlighting on a thematic front artificial intelligence (AI) in judicial decision-making in criminal matters; the co-dependency between cybercrime and organised crime; and data privacy concerns in relation to bring-your-own-device (BYOD) working practices, among other topics.

Chainalysis issues the 2023 cryptocurrency crime report

Private US company Chainalysis is a leading company in collecting and analyzing data used on cryptocurrency blockchains. In its annual report on cryptocurrency-related crime, they point out that illicit cryptocurrency volumes reach all-time highs amid a surge in sanctions and hacking. 

‘Overall, the share of all cryptocurrency activity associated with illicit activity has risen for the first time since 2019, from 0.12% in 2021 to 0.24% in 2022.’ The company assesses that an equivalent of $20.6B is used for illicit activities. 

A big part of that sum comes from the offenses related to the economic sanctions on Russia. This shows that a strict regime of sanctions is efficiently imposed on cryptocurrency exchanges, by the US department of the treasury, and international financial institutions. The report describes methods that are used for money laundering and fund transfers. As a key takeaway, Chainalisys points out that the impact of crypto sanctions depends on the jurisdiction and technical constraints.

Ransomware crypto payments

The report shows a decline in ransomware from 2021. Chainalisys claims that ransomware victims increasingly refuse to pay the ransom money hence pushing the criminals out of this scheme. The report is stating that “meaningful disruptions against ransomware actor groups are driving lower than expected successful extortion attempts”  In 2021, the US Office of Foreign Assets Control (OFAC) issued an advisory document about the risk of ‘sanction crimes’ that can rise from ransomware payments. OFAC advises all US companies to report ransomware to the FBI prior to any action. This is also considered to be one of the factors for the drop in ransomware payments. In addition, ransomware lifespan is significantly shorter. From 470 days in 2019, it is down to 70 days in 2022.

Money laundering

The report is stating a rise in money laundering activities from $14.2B in 2021 to $23.8B in 2022. The report is stating ‘underground money laundering services’ are a growing concern. Such groups use private channels on messaging apps to set and organise private transactions that are hard to track.

Cryptocurrency scams

Cryptocurrency scams and the use of cryptocurrency on darknet markets are on the decline compared to previous years.

Rise in ransomware attacks against manufacturing plans

A recent report by Dragos, a cybersecurity company, highlights the rise in ransomware attacks agains critical infrastructure and, in particular, against the manufacturing systems. The report shows that the manufacturing sector had at least 437 ransomware attacks in 2022, accounting for more than 70% of these disruptive attacks that industrial organisations had experienced the previous year.

The company identified a total of 605 ransomware attacks affecting the industrial sector in 2022, a 92% increase over the 315 attacks detected in 2021.

The report also epxlores the activity of two threat groups – Chernovite and Bentonite – that focus on attacking the industrial sector. While Chernovite targets electric, liquid, and natural gas companies in Europe and the USA, Bentonite mainly focuses on attacking maritime oil and gas companies, governments, and the manufacturing sector.