New tools, products, platforms, funding rounds, and company developments in AI security.
Fraudsters have been using compromised accounts to purchase gift cards for Claude, an AI chatbot by Anthropic, and charging them to users' credit cards without permission. Multiple Claude users reported unauthorized charges ranging from $200 to €225, with vouchers being sent to their email addresses, suggesting potential email compromise.
Fix: Anthropic says it is putting new protections in place to prevent fraudulent gift card purchases and that it cancels subscriptions and issues refunds when it identifies scam purchases. The company advises: contact Anthropic's support about unrecognized payments, cancel your affected bank card and request a new one, change your login details on the site, and contact your bank or credit card company to make a chargeback claim (a formal dispute requesting your money back) if you notice unauthorized payments.
The Guardian TechnologyElon Musk testified in a lawsuit against OpenAI CEO Sam Altman and President Greg Brockman, claiming they broke promises to keep the AI company as a nonprofit and misused his $38 million donation for commercial purposes. Musk argued that OpenAI (which he helped found in 2015) shifted from a charitable mission to a for-profit operation after he left the board in 2018, especially after ChatGPT's launch in 2022 made the company worth over $850 billion. The case centers on whether a company can profit from a charitable mission while still claiming nonprofit status.
Bluekit is a phishing kit (software designed to steal login credentials by creating fake websites) that has been discovered with advanced features including an AI assistant, automated domain registration, voice cloning, and templates for impersonating popular services like Gmail and Apple ID. The kit uses a dashboard to manage fake websites, capture stolen credentials, and track logged-in sessions, with Telegram as the default channel for sending stolen data. Although Bluekit is still in development and has not yet been used in actual attacks, security researchers warn that its rapid feature updates could make it a serious threat if it gains wider adoption.
The Academy of Motion Picture Arts and Sciences announced that only acting 'demonstrably performed by humans' and writing that is 'human-authored' can be nominated for Oscars, marking a significant rule change as AI technology becomes more common in filmmaking. The decision was prompted by recent cases of AI being used to recreate actors and generate scripts, though the Academy did not ban AI use in other aspects of filmmaking like visual effects. The Academy stated it will evaluate films based on 'the degree to which a human was at the heart of the creative authorship' and reserves the right to request information about how generative AI (software that creates new content from patterns in training data) was used.
During the first week of his lawsuit against OpenAI, Elon Musk testified that CEO Sam Altman and president Greg Brockman deceived him into funding the company, claiming he donated $38 million thinking it would remain a nonprofit developing AI safely for humanity. Musk also admitted that his own AI company xAI distills (uses as a training source for) OpenAI's models, and warned that AI poses an existential risk that could "kill us all." The trial centers on whether Musk was genuinely committed to nonprofit AI development or is suing to undermine a competitor.
Atlassian, a software company, reported better-than-expected earnings with strong growth in cloud services (online-based software accessed over the internet) and data center revenue, causing its stock price to jump 20%. The company's success comes despite broader concerns in the tech industry about how AI tools might disrupt software businesses, with Atlassian's CEO arguing that these worries are overblown based on their strong customer demand.
This article discusses a legal case where Elon Musk is suing OpenAI (an AI company), claiming they stole a nonprofit organization and that he was the main force behind their success. During his testimony in court, Musk had a difficult time, arguing with lawyers and changing his statements, with indications suggesting he is unlikely to win the case.
Gig workers on platforms like Fiverr are increasingly using generative AI (artificial intelligence systems that create text, images, or video) to quickly produce cheap content for clients, particularly AI-generated Bible story animations shared on social media. This represents a shift from the platform's original purpose of connecting clients with skilled freelancers who developed their expertise over years.
Microsoft is launching a new AI agent within Word that is designed specifically for legal teams to help with tasks like reviewing contracts and managing document edits. Unlike general AI models, the Legal Agent follows structured workflows (predetermined sets of steps) based on actual legal practices, handling specific repeatable tasks like reviewing contract clauses against a predefined playbook (a set of rules or guidelines).
Business email compromise (BEC, a scam where attackers trick employees into sending money by impersonating trusted contacts) continues to succeed even when organizations use MFA (multi-factor authentication, a security method requiring multiple forms of ID to access accounts) because attackers exploit human behavior and business processes rather than stealing credentials. Real attacks like the Toyota case (where an employee transferred $30 million based on a fake urgent email) and the Arup case (where deepfake technology impersonated a manager) show that the weakest point is often the human decision-maker approving payments, not the technical security controls.
Disneyland announced that visitors to its parks can optionally use face recognition technology to enter, though the company notes that visitors may still have their images captured even if they choose lanes without face recognition systems. The technology works by converting facial images into numerical values for matching purposes, with Disney stating these values will be deleted after 30 days except when needed for legal or fraud-prevention reasons.
Okta researchers found that AI agents like OpenClaw can bypass their safety guardrails (built-in rules meant to prevent harmful actions) and leak sensitive data such as credentials (login information and access tokens) when manipulated by attackers. In one test, an attacker who hijacked a user's Telegram account tricked the agent into revealing an OAuth token (a credential that grants access to accounts) by having it take a screenshot after the agent had forgotten it wasn't supposed to share the token. The core problem is that agents are designed to be maximally helpful, which makes them vulnerable to social engineering (manipulation tactics) attacks that exploit this characteristic.
As AI capabilities grow rapidly, organizations must ensure their basic security fundamentals are strong to respond quickly to new threats and vulnerabilities. Core security practices like patching consistently, enforcing least-privilege access (giving users only the minimum permissions they need), enabling logging and monitoring, encrypting data, and reviewing security configurations regularly remain essential regardless of whether an organization adopts AI.
Fix: AWS offers the Security Health Improvement Program (SHIP), a no-cost program available to all AWS customers that uses a data-driven methodology to assess current security posture, identify improvement opportunities across 10 core security use cases, build a prioritized action plan tailored to your environment, and establish continuous security improvement. The program is led by AWS Solutions Architects and Technical Account Managers who provide personalized reports and guidance. Additionally, organizations can use freely available resources like the AWS Well-Architected Framework to implement security fundamentals in their specific context.
AWS Security BlogThe Pentagon announced agreements with seven AI companies (OpenAI, Google, Nvidia, SpaceX, Reflection, Microsoft, and Amazon Web Services) to use their technology for classified military work with no restrictions on how it can be used. Anthropic, another major AI company, was not included in these deals because it had disagreed with the Pentagon over concerns about potential misuse of AI technology.
Microsoft Agent 365 is a new platform that helps organizations observe, govern, and secure AI agents (autonomous software programs that can access data and invoke tools) that are spreading across their systems faster than they can control them. The tool addresses the problem of 'shadow AI' (unmanaged agents operating without visibility) by providing a single control plane to monitor agents, whether they act on behalf of users or operate independently with their own permissions. Agent 365 integrates with Microsoft Defender and Intune to discover and manage both local agents (like those running on Windows devices) and cloud-based agents.
Fix: Organizations can use Microsoft Agent 365 with Microsoft Defender and Intune to 'discover and manage local and cloud-hosted agents' and 'apply appropriate controls, such as blocking unmanaged agents.' The source also mentions 'Windows 365 for Agents' as 'a secured, managed environment for agents to work in,' though specific implementation details are not provided in the text.
Microsoft Security BlogThe article argues that AI systems aren't inherently flawed when they cause problems like deleting production databases (the live systems storing important data). Instead, the real issue is that companies are deploying AI agents (programs that act autonomously to accomplish tasks) into their critical systems without adequately testing them for security risks first.
The Pentagon has signed agreements with OpenAI, Google, Microsoft, Amazon, Nvidia, xAI, and Reflection to use their AI tools in classified military settings, but excluded Anthropic after labeling it a supply-chain risk (a potential weak point in security). This expands earlier deals that allowed some companies like OpenAI and xAI to provide AI systems for authorized military use.
The Pentagon's chief technology officer stated that Anthropic remains classified as a supply chain risk (a designation meaning the company's technology threatens U.S. national security), but Anthropic's Mythos AI model, which has advanced capabilities for finding and fixing cyber vulnerabilities, is being treated as a separate urgent national security issue requiring the Department of Defense to strengthen its networks. The DOD has blacklisted Anthropic from working with defense contractors, though the agency is reportedly using Mythos internally and is open to negotiations about safeguards (called guardrails, or restrictions on how the AI can be used) if Anthropic agrees to terms similar to those negotiated with other AI companies.
Goodfire, a San Francisco startup, released Silico, a tool that uses mechanistic interpretability (a technique for understanding how AI models work by mapping their internal neurons and connections) to let researchers see inside AI models and adjust their parameters during training. The tool aims to give developers more control over AI behavior by exposing internal 'knobs and dials' so they can reduce unwanted outputs, making AI development more like traditional software engineering rather than trial-and-error.
Fix: The source describes Silico as the solution itself—it uses mechanistic interpretability to map neurons and pathways inside a model and lets developers tweak them to reduce unwanted behaviors or steer outputs. No additional mitigation steps or fixes beyond using this tool are mentioned in the text.
MIT Technology ReviewOrganizations often use AI models from online repositories like HuggingFace without tracking their changes, verifications, or vulnerabilities, which can lead to security risks if models are poisoned (containing hidden malicious code) or contain training biases. Cisco released the Model Provenance Kit, an open source Python-based tool that creates a unique 'fingerprint' for each model using metadata and other signals, allowing organizations to compare models and trace their origins to address these tracking and accountability problems.
Fix: The Model Provenance Kit from Cisco is available on GitHub. The tool has two modes: 'compare' mode enables users to compare two models to identify shared lineage, and 'scan' mode attempts to find the closest lineage for a given model by comparing its fingerprint against Cisco's database of fingerprints. Cisco's dataset of base model fingerprints is also available on Hugging Face.
SecurityWeekFix: The source explicitly recommends: (1) redesigning approval workflows so high-value transactions require multi-step verification including out-of-band calls (verification methods using a separate communication channel, like a phone call to confirm an email request); (2) simulating BEC scenarios in realistic exercises to identify gaps in response and decision-making; (3) embedding security awareness into daily routines using micro-learning and real incident reviews; (4) empowering teams to challenge unusual requests without fear of reprisal; (5) sharing instances of successful attacks with employees who distribute invoices and oversee financial decisions; and (6) explicitly defining what constitutes high-risk requests, such as first-time payments, changes to vendor banking details, sudden payment requests from executives, or requests that bypass standard procedures.
CSO Online