On Tuesday, Meta AI announced the development of Cicero, which it claims is the first AI to achieve human-level performance in the strategic board game Diplomacy. It’s a notable achievement because the game requires deep interpersonal negotiation skills, which implies that Cicero has obtained a certain mastery of language necessary to win the game.
[…]
Cicero learned its skills by playing an online version of Diplomacy on webDiplomacy.net. Over time, it became a master at the game, reportedly achieving “more than double the average score” of human players and ranking in the top 10 percent of people who played more than one game.
To create Cicero, Meta pulled together AI models for strategic reasoning (similar to AlphaGo) and natural language processing (similar to GPT-3) and rolled them into one agent. During each game, Cicero looks at the state of the game board and the conversation history and predicts how other players will act. It crafts a plan that it executes through a language model that can generate human-like dialogue, allowing it to coordinate with other players.
Enlarge/ A block diagram of Cicero, the Diplomacy-playing bot, provided by Meta.
Meta AI
Meta calls Cicero’s natural language skills a “controllable dialogue model,” which is where the heart of Cicero’s personality lies. Like GPT-3, Cicero pulls from a large corpus of Internet text scraped from the web. “To build a controllable dialogue model, we started with a 2.7 billion parameter BART-like language model pre-trained on text from the Internet and fine tuned on over 40,000 human games on webDiplomacy.net,” writes Meta.
The resulting model mastered the intricacies of a complex game. “Cicero can deduce, for example, that later in the game it will need the support of one particular player,” says Meta, “and then craft a strategy to win that person’s favor—and even recognize the risks and opportunities that that player sees from their particular point of view.”
Meta’s Cicero research appeared in the journal Science under the title, “Human-level play in the game of Diplomacy by combining language models with strategic reasoning.”
[…]
Meta provided a detailed site to explain how Cicero works and has also open-sourced Cicero’s code on GitHub. Online Diplomacy fans—and maybe even the rest of us—may need to watch out.
Mercedes is the latest manufacturer to lock auto features behind a subscription fee, with an upcoming “Acceleration Increase” add-on that lets drivers pay to access motor performance their vehicle is already capable of.
The $1,200 yearly subscription improves performance by boosting output from the motors by 20–24 percent, increasing torque, and shaving around 0.8 to 0.9 seconds off 0–60 mph acceleration when in Dynamic drive mode (via The Drive). The subscription doesn’t come with any physical hardware upgrades — instead, it simply unlocks the full capabilities of the vehicle, indicating that Mercedes intentionally limited performance to later sell as an optional extra. Acceleration Increase is only available for the Mercedes-EQ EQE and Mercedes-EQ EQS electric car models.
[…]
This comes just months after BMW sparked outrage by similarly charging an $18 monthly subscription in some countries for owners to use the heated seats already installed within its vehicles, just one of many features paywalled by the car manufacturer since 2020. BMW had previously also tried (and failed) to charge its customers $80 a month to access Apple CarPlay and Android Auto — features that other vehicle makers have included for free.
Unstable Diffusion is a server dedicated to the creation and sharing of AI generated NSFW.
We will seek to provide resources and mutual assistance to anyone attempting to make erotica, we will share prompts and artwork and tools specifically designed to get the most out of your generations, whether you’re using tools from the present or ones which may not have been invented as of this writing.
[…] Recently, the Microsoft Detection and Response Team (DART) has seen an increase in attackers utilizing token theft for this purpose. By compromising and replaying a token issued to an identity that has already completed multifactor authentication, the threat actor satisfies the validation of MFA and access is granted to organizational resources accordingly. This poses to be a concerning tactic for defenders because the expertise needed to compromise a token is very low, is hard to detect, and few organizations have token theft mitigations in their incident response plan.
[…]
Tokens are at the center of OAuth 2.0 identity platforms, such as Azure Active Directory (Azure AD). To access a resource (for example, a web application protected by Azure AD), a user must present a valid token. To obtain that token, the user must sign into Azure AD using their credentials. At that point, depending on policy, they may be required to complete MFA. The user then presents that token to the web application, which validates the token and allows the user access.
Figure 1. OAuth Token flow chart
When Azure AD issues a token, it contains information (claims) such as the username, source IP address, MFA, and more. It also includes any privilege a user has in Azure AD. If you sign in as a Global Administrator to your Azure AD tenant, then the token will reflect that. Two of the most common token theft techniques DART has observed have been through adversary-in-the-middle (AitM) frameworks or the utilization of commodity malware (which enables a ‘pass-the-cookie’ scenario).
[…]
When the user is phished, the malicious infrastructure captures both the credentials of the user, and the token.
If a regular user is phished and their token stolen, the attacker may attempt business email compromise (BEC) for financial gain.
[…]
A “pass-the-cookie” attack is a type of attack where an attacker can bypass authentication controls by compromising browser cookies.
[…]
Commodity credential theft malware like Emotet, Redline, IcedID, and more all have built-in functionality to extract and exfiltrate browser cookies. Additionally, the attacker does not have to know the compromised account password or even the email address for this to work— those details are held within the cookie.
[…]
Recommendations
Protect
Organizations can take a significant step toward reducing the risk of token theft by ensuring that they have full visibility of where and how their users are authenticating. To access critical applications like Exchange Online or SharePoint, the device used should be known by the organization. Utilizing compliance tools like Intune in combination with device based conditional access policies can help to keep devices up to date with patches, antivirus definitions, and EDR solutions. Allowing only known devices that adhere to Microsoft’s recommended security baselines helps mitigate the risk of commodity credential theft malware being able to compromise end user devices.
For those devices that remain unmanaged, consider utilizing session conditional access policies and other compensating controls to reduce the impact of token theft:
Reducing the lifetime of the session increases the number of times a user is forced to re-authenticate but minimizes the length of time session token is viable.
Reducing the viable time of a token forces threat actors to increase the frequency of token theft attempts which in turn provides defenders with additional chances at detection.
Plan and implement phishing resistant MFA solutions such as FIDO2 security keys, Windows Hello for Business, or certificate-based authentication for users.
While this may not be practical for all users, it should be considered for users of significant privilege like Global Admins or users of high-risk applications.
Users that hold a high level of privilege in the tenant should have a segregated cloud-only identity for all administrative activities, to reduce the attack surface from on-premises to cloud in the event of on-premises domain compromise and abuse of privilege. These identities should also not have a mailbox attached to them to prevent the likelihood of privileged account compromise via phishing techniques.
[…]
In instances of token theft, adversaries insert themselves in the middle of the trust chain and often subsequently circumvent security controls. Having visibility, alerting, insights, and a full understanding of where security controls are enforced is key. Treating both identity providers that generate access tokens and their associated privileged identities as critical assets is strongly encouraged.
