Entries Tagged "academic papers"

Page 1 of 71

Security Risks of Client-Side Scanning

Even before Apple made its announcement, law enforcement shifted their battle for backdoors to client-side scanning. The idea is that they wouldn’t touch the cryptography, but instead eavesdrop on communications and systems before encryption or after decryption. It’s not a cryptographic backdoor, but it’s still a backdoor — and brings with it all the insecurities of a backdoor.

I’m part of a group of cryptographers that has just published a paper discussing the security risks of such a system. (It’s substantially the same group that wrote a similar paper about key escrow in 1997, and other “exceptional access” proposals in 2015. We seem to have to do this every decade or so.) In our paper, we examine both the efficacy of such a system and its potential security failures, and conclude that it’s a really bad idea.

We had been working on the paper well before Apple’s announcement. And while we do talk about Apple’s system, our focus is really on the idea in general.

Ross Anderson wrote a blog post on the paper. (It’s always great when Ross writes something. It means I don’t have to.) So did Susan Landau. And there’s press coverage in the New York Times, the Guardian, Computer Weekly, the Financial Times, Forbes, El Pais (English translation), NRK (English translation), and — this is the best article of them all — the Register. See also this analysis of the law and politics of client-side scanning from last year.

Posted on October 15, 2021 at 9:30 AMView Comments

Recovering Real Faces from Face-Generation ML System

New paper: “This Person (Probably) Exists. Identity Membership Attacks Against GAN Generated Faces.

Abstract: Recently, generative adversarial networks (GANs) have achieved stunning realism, fooling even human observers. Indeed, the popular tongue-in-cheek website http://thispersondoesnotexist.com, taunts users with GAN generated images that seem too real to believe. On the other hand, GANs do leak information about their training data, as evidenced by membership attacks recently demonstrated in the literature. In this work, we challenge the assumption that GAN faces really are novel creations, by constructing a successful membership attack of a new kind. Unlike previous works, our attack can accurately discern samples sharing the same identity as training samples without being the same samples. We demonstrate the interest of our attack across several popular face datasets and GAN training procedures. Notably, we show that even in the presence of significant dataset diversity, an over represented person can pose a privacy concern.

News article. Slashdot post.

Posted on October 14, 2021 at 9:56 AMView Comments

Identifying Computer-Generated Faces

It’s the eyes:

The researchers note that in many cases, users can simply zoom in on the eyes of a person they suspect may not be real to spot the pupil irregularities. They also note that it would not be difficult to write software to spot such errors and for social media sites to use it to remove such content. Unfortunately, they also note that now that such irregularities have been identified, the people creating the fake pictures can simply add a feature to ensure the roundness of pupils.

And the arms race continues….

Research paper.

Posted on September 15, 2021 at 10:31 AMView Comments

Using “Master Faces” to Bypass Face-Recognition Authenticating Systems

Fascinating research: “Generating Master Faces for Dictionary Attacks with a Network-Assisted Latent Space Evolution.”

Abstract: A master face is a face image that passes face-based identity-authentication for a large portion of the population. These faces can be used to impersonate, with a high probability of success, any user, without having access to any user-information. We optimize these faces, by using an evolutionary algorithm in the latent embedding space of the StyleGAN face generator. Multiple evolutionary strategies are compared, and we propose a novel approach that employs a neural network in order to direct the search in the direction of promising samples, without adding fitness evaluations. The results we present demonstrate that it is possible to obtain a high coverage of the population (over 40%) with less than 10 master faces, for three leading deep face recognition systems.

Two good articles.

Posted on August 6, 2021 at 6:44 AMView Comments

Storing Encrypted Photos in Google’s Cloud

New paper: “Encrypted Cloud Photo Storage Using Google Photos.”

Abstract: Cloud photo services are widely used for persistent, convenient, and often free photo storage, which is especially useful for mobile devices. As users store more and more photos in the cloud, significant privacy concerns arise because even a single compromise of a user’s credentials give attackers unfettered access to all of the user’s photos. We have created Easy Secure Photos (ESP) to enable users to protect their photos on cloud photo services such as Google Photos. ESP introduces a new client-side encryption architecture that includes a novel format-preserving image encryption algorithm, an encrypted thumbnail display mechanism, and a usable key management system. ESP encrypts image data such that the result is still a standard format image like JPEG that is compatible with cloud photo services. ESP efficiently generates and displays encrypted thumbnails for fast and easy browsing of photo galleries from trusted user devices. ESP’s key management makes it simple to authorize multiple user devices to view encrypted image content via a process similar to device pairing, but using the cloud photo service as a QR code communication channel. We have implemented ESP in a popular Android photos app for use with Google Photos and demonstrate that it is easy to use and provides encryption functionality transparently to users, maintains good interactive performance and image quality while providing strong privacy guarantees, and retains the sharing and storage benefits of Google Photos without any changes to the cloud service

Posted on July 30, 2021 at 6:34 AMView Comments

Hiding Malware in ML Models

Interesting research: “EvilModel: Hiding Malware Inside of Neural Network Models.”

Abstract: Delivering malware covertly and detection-evadingly is critical to advanced malware campaigns. In this paper, we present a method that delivers malware covertly and detection-evadingly through neural network models. Neural network models are poorly explainable and have a good generalization ability. By embedding malware into the neurons, malware can be delivered covertly with minor or even no impact on the performance of neural networks. Meanwhile, since the structure of the neural network models remains unchanged, they can pass the security scan of antivirus engines. Experiments show that 36.9MB of malware can be embedded into a 178MB-AlexNet model within 1% accuracy loss, and no suspicious are raised by antivirus engines in VirusTotal, which verifies the feasibility of this method. With the widespread application of artificial intelligence, utilizing neural networks becomes a forwarding trend of malware. We hope this work could provide a referenceable scenario for the defense on neural network-assisted attacks.

News article.

Posted on July 27, 2021 at 6:25 AMView Comments

Insurance and Ransomware

As ransomware becomes more common, I’m seeing more discussions about the ethics of paying the ransom. Here’s one more contribution to that issue: a research paper that the insurance industry is hurting more than it’s helping.

However, the most pressing challenge currently facing the industry is ransomware. Although it is a societal problem, cyber insurers have received considerable criticism for facilitating ransom payments to cybercriminals. These add fuel to the fire by incentivising cybercriminals’ engagement in ransomware operations and enabling existing operators to invest in and expand their capabilities. Growing losses from ransomware attacks have also emphasised that the current reality is not sustainable for insurers either.

To overcome these challenges and champion the positive effects of cyber insurance, this paper calls for a series of interventions from government and industry. Some in the industry favour allowing the market to mature on its own, but it will not be possible to rely on changing market forces alone. To date, the UK government has taken a light-touch approach to the cyber insurance industry. With the market undergoing changes amid growing losses, more coordinated action by government and regulators is necessary to help the industry reach its full potential.

The interventions recommended here are still relatively light, and reflect the fact that cyber insurance is only a potential incentive for managing societal cyber risk.They include: developing guidance for minimum security standards for underwriting; expanding data collection and data sharing; mandating cyber insurance for government suppliers; and creating a new collaborative approach between insurers and intelligence and law enforcement agencies around ransomware.

Finally, although a well-functioning cyber insurance industry could improve cyber security practices on a societal scale, it is not a silver bullet for the cyber security challenge. It is important to remember that the primary purpose of cyber insurance is not to improve cyber security, but to transfer residual risk. As such, it should be one of many tools that governments and businesses can draw on to manage cyber risk more effectively.

Basically, the insurance industry incents companies to do the cheapest mitigation possible. Often, that’s paying the ransom.

News article.

Posted on July 1, 2021 at 11:01 AMView Comments

Intentional Flaw in GPRS Encryption Algorithm GEA-1

General Packet Radio Service (GPRS) is a mobile data standard that was widely used in the early 2000s. The first encryption algorithm for that standard was GEA-1, a stream cipher built on three linear-feedback shift registers and a non-linear combining function. Although the algorithm has a 64-bit key, the effective key length is only 40 bits, due to “an exceptional interaction of the deployed LFSRs and the key initialization, which is highly unlikely to occur by chance.”

GEA-1 was designed by the European Telecommunications Standards Institute in 1998. ETSI was — and maybe still is — under the auspices of SOGIS: the Senior Officials Group, Information Systems Security. That’s basically the intelligence agencies of the EU countries.

Details are in the paper: “Cryptanalysis of the GPRS Encryption Algorithms GEA-1 and GEA-2.” GEA-2 does not have the same flaw, although the researchers found a practical attack with enough keystream.

Hacker News thread.

EDITED TO ADD (6/18): News article.

Posted on June 17, 2021 at 1:51 PMView Comments

1 2 3 71

Sidebar photo of Bruce Schneier by Joe MacInnis.