Coverage: Last 24 hours
Today’s Highlights
Multiple significant vulnerabilities and trust failures emerged around AI gateways and Microsoft Copilot, highlighting real-world privilege escalation and data exfiltration risks for defenders managing large-scale SaaS and AI deployments. Key themes today include ongoing AI supply chain risk, toxic privilege escalation, gaps in transparency for both government AI and corporate SaaS, and the need for rapid technical and policy updates in response to deepfake abuse and legacy software exploitation.
Table of Contents
- DOJ seizes CFAKE, SOCFAKE deepfake nude sites under TAKE IT DOWN Act
- ⚡ Weekly Recap: Chrome 0-Day, UniFi Exploits, macOS Stealers, VPN Flaw and More
- LiteLLM Vulnerability Chain Lets Low-Privilege Users Take Over AI Gateway Servers
- One-Click Microsoft 365 Copilot Flaw Could Have Let Attackers Steal Emails, Files, and MFA Codes
- AI could help win ‘race against extinction’ of vital plants, say botanists
- AI use by the US government is ballooning. And the lack of transparency is troubling | Nathan E Sanders and Bruce Schneier
- Why do South Koreans love AI so much?
- ‘Pretty Crazy’ Token Usage Is Testing Bosses’ Bet on AI
- Anthropic Is Still at Odds With the White House Over Claude Fable 5
- Meta CTO Andrew Bosworth Admits the Company’s AI Reorg Was ‘Atrocious’
Top Stories
DOJ seizes CFAKE, SOCFAKE deepfake nude sites under TAKE IT DOWN Act
Source: BleepingComputer | Risk: Medium | Impacted: Platforms hosting AI-generated content, Cloud storage operators, Legal and compliance teams
Summary: The U.S. Department of Justice announced Friday that it has seized the CFAKE.com and SOCFAKE.com websites, which allegedly hosted nonconsensual AI-generated nude images and videos of women, in what appears to be the first publicly announced domain seizure under the TAKE IT DOWN Act.
Why it matters: New legal action against nonconsensual deepfake content hosting signals greater enforcement risk for platforms handling AI-generated media that could cross into privacy or unlawful use cases.
Practitioner Perspective
The DOJ’s takedown of CFAKE and SOCFAKE under recent legislation should prompt platforms and cloud providers hosting user-generated content to reevaluate their risk exposure. Tools and APIs for AI-driven image and video generation are increasingly being abused for unlawful and reputation-damaging purposes. Security, privacy, and legal teams must coordinate to proactively detect, prevent, and remove flagged AI content before law enforcement intervention forces their hand or exposes the business to regulatory fines. Consider that abuse reporting and monitoring processes may not be keeping pace with generative AI risks, which are evolving faster than most existing policy frameworks. The critical work is linking technical detection with upstream policy enforcement for these new classes of content.
Recommended Actions
- Review and update abuse monitoring for AI-generated images and videos, specifically targeting deepfake nude content
- Collaborate with legal to ensure content moderation policies align with the TAKE IT DOWN Act and similar laws
Emerging Signals
(No stories today)
Exploits & CVEs
⚡ Weekly Recap: Chrome 0-Day, UniFi Exploits, macOS Stealers, VPN Flaw and More
Source: The Hacker News | Risk: High | Impacted: Organizations with legacy software, Environments running UniFi, Chrome, or VPN appliances, IT operations teams
Summary: Stuff broke again. Not in a movie way. An old tool was left exposed. An abandoned package was abused. A deprecated feature was still running in prod. This week is the same lesson in a new form: phishing kits are easier to rent, AI names are useful bait, old login paths still fail, and forgotten software keeps becoming someone else’s.
Why it matters: A surge in opportunistic and targeted exploitation of unmaintained software, abandoned packages, and deprecated features leads to persistent initial access and lateral movement risks, particularly where IT hygiene lags behind attacker automation.
Practitioner Perspective
In the past week, attacks have targeted Chrome browsers, UniFi devices, VPN appliances, and both widely deployed and legacy endpoints, exploiting the full stack from aging login paths to unpatched 0-days. Attackers are increasingly automating their exploitation of ‘forgotten’ IT assets, these often evade vulnerability management because they’re considered out of scope or ‘minor’. Even commodity phishing kits now weaponize old software and authentication paths, leveraging the broad exposure window defenders permit by delaying removal or update. Dedicate cycles to asset and software discovery: your highest risk is what you no longer account for. Old and abandoned software needs to be treated as a high-priority removal, not just a patching afterthought.
Recommended Actions
- Deploy the latest Chrome patches immediately, recent 0-days have been actively exploited in the wild
- Prioritize updating or isolating UniFi device firmware, focusing on models with known authentication or RCE flaws
AI Security
LiteLLM Vulnerability Chain Lets Low-Privilege Users Take Over AI Gateway Servers
Source: The Hacker News | Risk: Critical | Impacted: AI gateway operators, Organizations brokering AI model access, Development teams using LiteLLM, Secrets management platforms
Summary: A default low-privilege account on a LiteLLM proxy can climb to full admin and run code on the server by chaining three vulnerabilities, researchers at Obsidian Security disclosed LiteLLM is a widely deployed open-source AI gateway that brokers calls to more than 100 model providers behind one OpenAI-compatible interface. A server takeover exposes every provider key it holds, the secrets
Why it matters: A privilege escalation chain allows an attacker with minimal rights to take over the gateway and exfiltrate all managed provider keys, placing every connected AI model and downstream application at risk of compromise.
Practitioner Perspective
Organizations using LiteLLM as a broker between multiple AI model providers face the prospect of a single low-privilege user leading to full environment compromise. This mirrors classic proxy and privilege escalation patterns, but now in a context where stolen secrets can cross many internal and external trust boundaries. Your incident response and detection coverage for these gateways are likely nascent, and most asset inventories will underestimate the spread of exposed secrets if a server is pwned. Any service or app wired to LiteLLM could be exposed, not just the server itself. The most urgent task is to enumerate all places this gateway runs and apply available patches or mitigations, while preparing for potential credential rotation across every provider key integrated.
Recommended Actions
- Enumerate all LiteLLM deployments, both production and test infrastructure, and review for default or low-privilege accounts
- Apply official patches or configuration mitigations for the LiteLLM privilege escalation chain immediately
One-Click Microsoft 365 Copilot Flaw Could Have Let Attackers Steal Emails, Files, and MFA Codes
Source: The Hacker News | Risk: High | Impacted: M365 Copilot users, Organizations using Microsoft Enterprise Search, SaaS security operations teams, Cloud identity administrators
Summary: A single click on a trusted Microsoft link could have let an attacker pull emails, calendar details, and indexed files out of Microsoft 365 Copilot Enterprise Search. Researchers at Varonis Threat Labs chained three bugs into a one-click exfiltration path they call SearchLeak. Because the link pointed to a real microsoft.com domain, traditional anti-phishing and URL filtering tools were
Why it matters: Credential and data exfiltration could occur even with robust phishing defense in place, since the exploit leverages a legitimate Microsoft domain and subverts typical anti-phishing and URL filtering controls.
Practitioner Perspective
M365 tenants that have enabled Copilot or Enterprise Search features are exposed to a situation where a single victim click can leak sensitive data and MFA codes to an attacker. This demonstrates that even the most reputable SaaS platforms can introduce gaps that bypass established layered defenses. Posture management and incident response playbooks must catch these edge cases, as traditional user awareness training and gateway filtering will not be enough. Review tenant logs for use of the SearchLeak chain, and coordinate with identity and SaaS app teams to assess blast radius where Copilot has been widely piloted or adopted. The key takeaway: do not treat trusted SaaS domains as inherently safe for all internal data flows.
Recommended Actions
- Identify and inventory all users and groups with Microsoft Copilot enabled across your O365 tenant
- Search M365 audit and sign-in logs for anomalous data access patterns related to Copilot and Enterprise Search
AI could help win ‘race against extinction’ of vital plants, say botanists
Source: The Guardian | Risk: Low | Impacted: Conservation researchers, Environmental agencies
Summary: Tech is helping to identify and save new specimens and could open ‘genomic goldmine’ of fungi data The rise of AI and digitisation could be a turning point in the “race against extinction” faced by botanists trying to identify and save vital plants before they vanish, according to a major report from Royal Botanic Gardens, Kew. New technology is enabling
Why it matters: The accelerated identification of vulnerable species and critical ecosystems can enhance global conservation efforts, but also raises long-term considerations around sensitive ecological data protection.
Practitioner Perspective
As conservation organizations leverage AI for specimen identification and gene sequencing, they must be mindful of how this sensitive ecological intelligence could be valued or targeted by interested third parties. Data on rare or endangered species, if mishandled, can attract risk to those species themselves. Building secure, access-controlled systems for ecological datasets is as important as leveraging the latest analytics.
Recommended Actions
- Inventory AI tools and data storage solutions for ecological research; assess access controls
- Regularly review compliance with cross-border and indigenous data protection guidelines
AI use by the US government is ballooning. And the lack of transparency is troubling | Nathan E Sanders and Bruce Schneier
Source: The Guardian | Risk: Medium | Impacted: Government agency infosec teams, Regulated industry compliance leads, Data governance officers
Summary: The list of government AI use cases has ballooned by 70% since Biden left office and includes many plans to hand over sensitive governmental functions to AI On 14 April, the Trump administration quietly acknowledged the widespread use of AI to automate government processes. The office of management and budget (OMB) disclosed a staggering 3,611 active or planned use cases
Why it matters: Dramatic increase in federal AI adoption without robust disclosure creates the potential for damage to data sovereignty, loss of auditability, and misalignment with agency compliance requirements.
Practitioner Perspective
The scale and opacity of AI use in the US government should serve as a wakeup call for private sector organizations mirroring this trend without rigorous inventory, oversight, or risk review. Relying on machine-managed workflows or data processing may unknowingly create audit gaps or introduce unknown outside code into sensitive processes. Many agencies and private enterprises likely do not know the full extent of AI application’s privilege or scope within their environment. The key is continuous mapping, not just point-in-time review, of where and how AI is embedded in systems handling sensitive, regulated, or mission-critical data. The top-priority is building an accessible and authoritative inventory of every sanctioned AI workflow, especially those automating core processes.
Recommended Actions
- Create and maintain a centralized list of all AI-driven workflows and tools in your environment
- Perform targeted risk and compliance reviews for workflows automating functions previously handled by humans
Why do South Koreans love AI so much?
Source: MIT Tech Review AI | Risk: Low | Impacted: General public, Policy analysts
Summary: This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. When I landed in Seoul after a grueling 12-hour flight from San Francisco, I walked through an unmanned immigration checkpoint, where a machine scanned my face and passport. On the subway home, …
Why it matters: Rapid adoption and public trust in AI for daily life can offer lessons about sociotechnical acceptance and scaling, while highlighting the privacy implications of widespread biometric systems.
Practitioner Perspective
Widespread public support for automated checkpoints and AI services in South Korea offers a glimpse into the future for many regions, but it also intensifies scrutiny around centralized biometric processing. Privacy and ethics teams should observe how societal expectations shift with rapid technology adoption, and prepare for corresponding regulatory changes.
Recommended Actions
- Monitor global regulatory responses to biometric data collection and AI-centric public infrastructure
- Evaluate internal policies for cross-border data transfers and public-facing AI deployments
‘Pretty Crazy’ Token Usage Is Testing Bosses’ Bet on AI
Source: The Verge AI | Risk: Low | Impacted: CTOs, Procurement teams, Cloud cost controllers
Summary: A Silicon Valley software maker and an ecommerce company reveal to WIRED how they are navigating the emerging challenge of “tokenomics.”
Why it matters: Organizations are beginning to confront the hidden operational and financial impacts that high-volume AI model usage can create, as unexpected compute costs mount.
Practitioner Perspective
The sudden appearance of massive token compute bills signals a need for improved monitoring of generative AI usage and cost controls in cloud platforms. Unplanned scale-ups can outpace internal billing, review, or security alerting processes. Consider enabling granular metering and alerts as part of cloud AI deployments.
Recommended Actions
- Apply resource-level billing and tracking for all AI model API consumption
- Set up usage thresholds and cost overrun alerts in primary cloud and SaaS providers
Anthropic Is Still at Odds With the White House Over Claude Fable 5
Source: The Verge AI | Risk: Medium | Impacted: AI policy teams, Risk analysts
Summary: Anthropic leaders flew to Washington, DC, to meet with White House officials on Monday. After high-level talks, they’re still split on the risk Claude Fable 5 presents.
Why it matters: Persistent gaps in understanding large model risk, especially at executive or regulatory levels, impede progress toward effective guardrails, incident investigation standards, and public trust.
Practitioner Perspective
Major disagreements between model builders and governments over what constitutes ‘acceptable risk’ could slow deployment of both technical and policy solutions for model monitoring. Risk and policy teams must press for detailed model disclosures from vendors and support independent validation.
Recommended Actions
- Request full documentation and change logs from AI vendors for each new model release
- Support robust internal and third-party model risk reviews prior to deployment
Meta CTO Andrew Bosworth Admits the Company’s AI Reorg Was ‘Atrocious’
Source: The Verge AI | Risk: Low | Impacted: Meta employees, Tech sector HR
Summary: In an internal memo seen by WIRED, Bosworth promised employees more stability, better communication, and the return of workplace perks as the company seeks to improve morale.
Why it matters: Organizational turbulence in a major AI player can introduce unpredictability and morale risks that may affect project velocity and alignment across large tech ecosystems.
Practitioner Perspective
AI and security leaders in large companies should track whether organizational restructuring affects technical debt, code review cycles, or incident response capacity. Proactive team communications and early detection of morale dips can mitigate operational degradation.
Recommended Actions
- Conduct regular pulse surveys and feedback loops post-restructuring
- Assign technical stewards to ensure code and process continuity amid team churn
Defensive Actions
- Enumerate all LiteLLM deployments, both production and test infrastructure, and review for default or low-privilege accounts
- Apply official patches or configuration mitigations for the LiteLLM privilege escalation chain immediately
- Audit LiteLLM access logs for suspicious privilege elevation or atypical code execution events
- Rotate all provider keys stored in compromised or unpatched LiteLLM instances
- Update threat modeling for systems integrated via LiteLLM to account for full environment exposure if the broker is breached
- Identify and inventory all users and groups with Microsoft Copilot enabled across your O365 tenant
- Search M365 audit and sign-in logs for anomalous data access patterns related to Copilot and Enterprise Search
- Monitor for clicks and downstream activity involving legitimate microsoft.com links that led to large file, mailbox, or MFA code access
- Coordinate with Microsoft for additional tenant telemetry or detection guidance specific to the SearchLeak attack chain
- Review and restrict unnecessary Copilot deployments while confirming least-privilege and conditional access are enforced
- Deploy the latest Chrome patches immediately, recent 0-days have been actively exploited in the wild
- Prioritize updating or isolating UniFi device firmware, focusing on models with known authentication or RCE flaws
What We’re Watching
- Ongoing impact and detection gaps from cross-cloud AI broker vulnerabilities
- Knock-on effect of legal action against AI-generated deepfake content for global platform governance
- Internal security monitoring best practices as SaaS, AI, and legacy system risks continue to converge
- Policy and compliance changes tied to increased federal and corporate use of AI, especially around transparency and auditability
- AI adoption and ethical debates evolving rapidly in both public-sector and multinational tech environments
Categories: Artificial Intelligence, Cybersecurity Blog
TECHMANIACS.com 
Leave a comment