All tracked items across vulnerabilities, news, research, incidents, and regulatory updates.
Google DeepMind's leader Sir Demis Hassabis told the BBC that more research is urgently needed to address AI threats, particularly the risk of bad actors misusing the technology and losing control of increasingly powerful autonomous systems (software that makes decisions without human input). While tech leaders and most countries at the AI Impact Summit called for stronger global governance and "smart regulation" of AI, the US rejected this approach, arguing that excessive rules would slow progress.
This article discusses passwordless authentication, an alternative to traditional passwords that uses standards like FIDO2 and Passkeys (cryptographic keys stored on devices instead of passwords) to improve security and reduce administrative burden. The article explains that the FIDO Alliance manages these standards and lists ten commercial passwordless solutions from vendors like AuthID, Axiad, Beyond Identity, and CyberArk that offer features such as biometric authentication, risk-based evaluation of login attempts, and integration with existing identity management systems.
Nvidia is in talks to invest up to $30 billion in OpenAI as part of a funding round that could value the AI startup at $730 billion, separate from a previously announced $100 billion infrastructure agreement. This new investment is not tied to any specific deployment milestones, and the deal is still under negotiation with details subject to change.
Google released Gemini Pro 3.1, a new large language model (LLM, an AI trained on vast amounts of text to understand and generate language), which achieved record scores on independent performance benchmarks like Humanity's Last Exam and APEX-Agents. The model is currently in preview and represents a major improvement over the previous Gemini 3 version, particularly for agentic work (tasks where the AI breaks down complex problems into multiple steps and executes them).
OpenClaw is a personal AI assistant with a macOS desktop client that can be triggered through deep links (special URLs that open apps). In versions 2026.2.6 through 2026.2.13, attackers could hide malicious commands by padding messages with whitespace, so users would see only a harmless preview but the full hidden command would execute when they clicked 'Run'. This works because the app only displayed the first 240 characters in the confirmation dialog before executing the entire message.
SillyTavern is a locally installed interface for interacting with text generation AI models and other AI tools. Versions before 1.16.0 had an SSRF vulnerability (server-side request forgery, where an attacker can make the server send requests to internal networks or services it shouldn't access), allowing authenticated users to read responses from internal services and private network resources through the asset download feature.
YouTube is expanding its conversational AI tool to smart TVs, gaming consoles, and streaming devices, allowing users to ask questions about video content using an 'Ask' button or voice commands without pausing playback. The feature, currently available to select users over 18 in five languages, lets viewers get instant answers about things like recipe ingredients or song background information. This expansion reflects YouTube's growing dominance in TV viewing, with competitors like Amazon, Roku, and Netflix also developing their own conversational AI features for television.
OpenClaw, an npm package, used SHA-1 (an outdated hashing algorithm with known weaknesses) to create identifiers for Docker and browser sandbox configurations. An attacker could exploit hash collisions (two different configurations producing the same hash) to trick the system into reusing the wrong sandbox, leading to cache poisoning (corrupting stored data) and unsafe sandbox reuse.
Microsoft's Semantic Kernel Python SDK has an RCE vulnerability (remote code execution, where an attacker can run commands on a system they don't own) in the `InMemoryVectorStore` filter functionality, which allows attackers to execute arbitrary code. The vulnerability affects the library used for building AI applications with vector storage (a database that stores AI embeddings, which are numerical representations of data).
India is hosting a major AI Impact Summit attracting executives from major AI companies and tech firms to drive investment and innovation in the country. The event showcases significant AI development activity, including new investments in Indian AI startups, partnerships between international AI companies and Indian firms, and announcements of local AI infrastructure projects, while also highlighting concerns about AI's potential impact on traditional IT services jobs.
Google released Gemini 3.1 Pro on February 19, 2026, a new AI model priced at half the cost of Claude Opus 4.6 with similar performance benchmarks. The model shows improved ability to generate SVG animations (scalable vector graphics, images made from code rather than pixels) compared to its predecessor, though it is currently experiencing slow response times and occasional errors due to high demand at launch.
PromptSpy is Android malware that uses Google's Gemini AI chatbot to maintain persistence on infected devices by sending UI information to Gemini, which then instructs the malware where to tap or swipe to add itself to recent apps. The malware also abuses Accessibility Services (a system feature that allows apps to interact with the device interface) to prevent users from uninstalling it by overlaying invisible blocks over removal buttons.
Fix: According to ESET researchers, victims can remove PromptSpy by rebooting the device into Safe Mode, where third-party apps are disabled and can be uninstalled normally.
SecurityWeekThe Electronic Frontier Foundation (EFF) introduced a policy for open-source contributions that requires developers to understand any code they submit and to write comments and documentation themselves, even if they use LLMs (large language models, AI systems trained to generate human-like text) to help. While the EFF does not completely ban LLM-assisted code, they require disclosure of LLM use because AI-generated code can contain hidden bugs that scale poorly and create extra work for reviewers, especially in under-resourced teams.
Fix: The source explicitly states that contributors must disclose when they use LLM tools. The EFF's policy requires that: (1) contributors understand the code they submit, and (2) comments and documentation be authored by a human rather than generated by an LLM. No technical patch, update, or automated mitigation is discussed in the source.
EFF Deeplinks BlogRoundCube Webmail has a deserialization of untrusted data vulnerability (a flaw where the program unsafely processes data from users, which can be exploited to run malicious code) in its settings upload feature because a URL parameter called _from is not properly checked. This allows authenticated users (those who have logged in) to execute remote code execution (run commands on the server without owning it), and it is currently being exploited by attackers in real-world attacks.
Fix: Apply security updates to RoundCube Webmail version 1.6.11 or version 1.5.10, according to vendor instructions at https://roundcube.net/news/2025/06/01/security-updates-1.6.11-and-1.5.10. Alternatively, follow applicable BOD 22-01 guidance for cloud services or discontinue use of the product if mitigations are unavailable.
CISA Known Exploited VulnerabilitiesRoundCube Webmail has a cross-site scripting vulnerability (XSS, a type of attack where malicious code is injected into a webpage to run in users' browsers) that can be triggered through the animate tag in SVG documents. This vulnerability is currently being actively exploited by attackers in the wild. Organizations using RoundCube Webmail need to take action by the March 13, 2026 deadline.
Fix: Apply mitigations per vendor instructions, follow applicable BOD 22-01 guidance for cloud services, or discontinue use of the product if mitigations are unavailable. Security updates are available in versions 1.6.12 and 1.5.12 (see vendor release notes at https://roundcube.net/news/2025/12/13/security-updates-1.6.12-and-1.5.12).
CISA Known Exploited VulnerabilitiesFix: The issue is fixed in version 2026.2.14. The source also mentions mitigations: do not approve unexpected 'Run OpenClaw agent?' prompts triggered while browsing untrusted websites, and use deep links only with a valid authentication key for trusted personal automations.
NVD/CVE DatabaseResearchers discovered PromptSpy, the first known Android malware that uses generative AI (specifically Google's Gemini model) during its operation to help it persist on infected devices by adapting how it locks itself in the Recent Apps list across different Android manufacturers. Beyond this AI feature, PromptSpy functions as spyware with a VNC module (remote access tool) that allows attackers to view and control the device, intercept passwords, record screens, and capture installed apps. The malware also uses invisible UI overlays to block users from uninstalling it or disabling its permissions.
Fix: According to ESET, victims must reboot into Android Safe Mode so that third-party apps are disabled and cannot block the malware's uninstall.
BleepingComputerNIST announced the AI Agent Standards Initiative to develop standards and safeguards for agentic AI (autonomous AI systems that can perform tasks independently), with the goal of building public confidence and ensuring safe adoption. The initiative faces criticism for moving too slowly, as real-world security incidents involving agentic AI (like the EchoLeak vulnerability in Microsoft 365 Copilot and the OpenClaw agent that can let attackers access user data) are already occurring faster than standards can be developed.
Fix: The vulnerability has been patched in version 1.16.0 by introducing a whitelist domain check for asset download requests. It can be reviewed and customized by editing the `whitelistImportDomains` array in the `config.yaml` file.
NVD/CVE DatabaseFix: Update to version 2026.2.15 or later. The fix replaces SHA-1 with SHA-256 (a stronger hashing algorithm with better collision resistance) for generating these sandbox identifiers.
GitHub Advisory DatabaseFix: Upgrade to python-1.39.4 or higher. As a temporary workaround, avoid using `InMemoryVectorStore` for production scenarios.
GitHub Advisory DatabaseA hacker exploited a vulnerability in Cline, an open-source AI coding agent, to trick it into installing OpenClaw (a viral AI agent that can perform autonomous actions) across many systems. The vulnerability allowed attackers to use prompt injection (hidden malicious instructions embedded in input) to make Claude, the AI powering Cline, execute unintended commands, highlighting growing security risks as more people deploy autonomous software.
Microsoft 365 Copilot Chat, an AI work assistant, had a bug that caused it to accidentally access and summarize confidential emails from users' draft and sent folders, even though those emails were marked as confidential and protected by security policies. The issue affected enterprise users and was first discovered in January, though Microsoft says no one gained access to information they weren't already authorized to see. Microsoft has since rolled out a configuration update worldwide to fix the problem.
Fix: Microsoft has rolled out a configuration update to fix the issue. According to a Microsoft spokesperson: 'A configuration update has been deployed worldwide for enterprise customers.'
BBC TechnologyPromptSpy is Android malware that uses Gemini (Google's AI chatbot) to automatically keep itself running on victims' devices by analyzing the screen and sending instructions on how to stay in the recent apps list. The malware also uses accessibility services (special permissions that let apps control your device without user input) to steal data, prevent uninstallation, and give attackers remote access through a VNC module (virtual network computing, software for controlling devices remotely), and it's being distributed through fake websites targeting users in Argentina.