In this episode for April 23, 2026, Jeremy explores a week where "first principles" in security are being forgotten in the rush to adopt AI. From guessable API endpoints exposing Anthropic’s most powerful model to a $10,000 fine for a lawyer’s AI "slop," the message of the week is clear: There is no AI without API security.

Key Stories & Developments:

• The Mythos API Leak: Unauthorized actors gained access to Anthropic’s Claude Mythos model by simply guessing API naming conventions. This classic case of Broken Function Level Authorization highlights a major oversight in the rollout of sensitive models.
• Shadow AI Agents: A new survey from the Cloud Security Alliance reveals that 82% of enterprises have unknown AI agents operating without security oversight.
• The $10K Hallucination: An Oregon lawyer was fined $10,000 for "AI slop" in court filings, setting a firm legal precedent that AI error does not excuse professional negligence.
• MCP Design Flaws: The Model Context Protocol (MCP), designed to wrap APIs in human language, is proving vulnerable to coercion. Attackers are using human language requests to probe back-end systems through NGINX.
• "Logjack": New research into "Logjack" shows how malicious prompts hidden in system logs can compromise the LLMs used to analyze them.
• Meta Keystroke Capturing: Reports indicate Meta is capturing employee keystrokes to refine internal AI training sets, raising massive concerns about insider risk and password exfiltration.

Shadow AI agents are the new Shadow IT. Are you part of the 82% with zero visibility into your AI agents? Discover every agent and API connection in 15 minutes. Book your FireTail demo: https://www.firetail.ai/schedule-your-demo

Episode Links

https://www.inc.com/kevin-haynes/faulty-ai-leads-to-record-10000-fine-for-oregon-lawyer/91322007

https://www.nytimes.com/2026/04/17/us/oregon-winery-ai-legal-fight.html

https://techcrunch.com/2026/04/21/meta-will-record-employees-keystrokes-and-use-it-to-train-its-ai-models/

https://cloudsecurityalliance.org/press-releases/2026/04/21/new-cloud-security-alliance-survey-reveals-82-of-enterprises-have-unknown-ai-agents-in-their-environments

https://techcrunch.com/2026/04/20/app-host-vercel-confirms-security-incident-says-customer-data-was-stolen-via-breach-at-context-ai/

https://www.securityweek.com/by-design-flaw-in-mcp-could-enable-widespread-ai-supply-chain-attacks/

https://www.theregister.com/2026/04/16/anthropic_mcp_design_flaw/

https://www.darkreading.com/application-security/critical-mcp-integration-flaw-nginx-risk

https://www.helpnetsecurity.com/2026/04/16/llm-router-security-risk-agent-commands/

https://oddguan.com/blog/comment-and-control-prompt-injection-credential-theft-claude-code-gemini-cli-github-copilot/

https://arxiv.org/abs/2604.15368

https://venturebeat.com/security/microsoft-salesforce-copilot-agentforce-prompt-injection-cve-agent-remediation-playbook

https://techcrunch.com/2026/04/21/unauthorized-group-has-gained-access-to-anthropics-exclusive-cyber-tool-mythos-report-claims/

https://aisle.com/blog/ai-cybersecurity-after-mythos-the-jagged-frontier

https://www.darkreading.com/vulnerabilities-threats/every-old-vulnerability-ai-vulnerability

https://www.theregister.com/2026/04/20/lovable_denies_data_leak/

This week, Jeremy breaks down a sophisticated bypass of Apple Intelligence and explores a hardware-level GPU threat that turns "vandalism" into full system takeovers. We also look at the massive data fallout from the Mercor supply chain breach and why "Claude Mythos" is officially ending the era of slow vulnerability management.

Key Stories & Developments:

• NeuralExec vs. Apple: Researchers reveal a 76% success rate in bypassing Apple Intelligence safety filters using Right-to-Left (RTL) Unicode overrides.
• The 4TB Mercor Leak: The fallout from the LiteLLM supply chain attack is confirmed: 4 terabytes of data stolen, leading Meta to pause contracts and OpenAI to investigate exposure.
• GPU-Breach: A new technique from the University of Toronto moves beyond "bit-flipping" to gain God-mode over GPU memory, threatening cryptographic secrets.
• Secret Sprawl Explosion: GitGuardian reports a 34% jump in exposed secrets, with AI service credentials (like OpenRouter and Google API keys) being the fastest-growing category.
• The Death of the Patch Cycle: "Claude Mythos" has flipped the script—99% of its AI-discovered zero-days are now valid, forcing a realization that this is no longer an AI security problem, but a high-speed vulnerability management crisis.

Episode Links

1. https://9to5mac.com/2026/04/09/researchers-detail-how-a-prompt-injection-attack-bypassed-apple-intelligence-protections/
2. https://securityboulevard.com/2026/04/bypassing-llm-supervisor-agents-through-indirect-prompt-injection/
3. https://cybersecurityjournal.ca/techtalk/83883-flowise-cve-2025-59528-rce-exploitation-ai-agent-builder-2026-04-08/
4. https://cyberscoop.com/grafanaghost-grafana-prompt-injection-vulnerability-data-exfiltration/
5. https://techcrunch.com/2026/04/09/after-data-breach-10b-valued-startup-mercor-is-having-a-month/
6. https://www.helpnetsecurity.com/2026/04/14/gitguardian-ai-agents-credentials-leak/
7. https://securityaffairs.com/190455/security/gpubreach-exploit-uses-gpu-memory-bit-flips-to-achieve-full-system-takeover.html
8. https://aisle.com/blog/system-over-model-zero-day-discovery-at-the-jagged-frontier
9. https://openai.com/index/scaling-trusted-access-for-cyber-defense/
10. https://www.npr.org/2026/04/11/nx-s1-5778508/anthropic-project-glasswing-ai-cybersecurity-mythos-preview
11. https://labs.cloudsecurityalliance.org/wp-content/uploads/2026/04/mythosready.pdf
12. https://www.businessinsider.com/andon-market-luna-ai-agent-managed-store-san-francisco-2026-4#

Worried about AI security?

Get Complete AI Visibility in 15 Minutes. Discover all of your shadow AI now. Book a demo of Firetail's AI Security & Governance Platform: https://www.firetail.ai/request-a-demo

In this episode for April 9, 2026, Jeremy covers a week dominated by highly sophisticated supply chain attacks and the emergence of "Project Glasswing", an internal Anthropic project revealing that next-gen AI models may be "too good" at finding zero-day vulnerabilities.

Key Stories & Developments:

• The FBI's IC3 Report: For the first time in 25 years, the FBI has specifically categorized AI-enabled fraud, which accounted for $893 million in losses across BEC, romance, and investment scams.
• Ollama Exposure Spikes: A Shodan scan reveals that publicly exposed Ollama instances have jumped from 1,100 in September 2025 to over 25,000 in April 2026.
• Critical Infrastructure CVEs: Both MLflow and PraisonAI received maximum CVSS scores of 10.0 for flaws allowing unauthenticated code execution and command injection.
• The Axios Supply Chain Heist: In a sophisticated "long con," threat actors (Team PCP) spent weeks building rapport with the Axios project maintainer via a fake Slack workspace. They eventually lured the maintainer into downloading malware, allowing them to inject a Remote Access Trojan (RAT) into a package installed 600,000 times.
• Project Glasswing (Claude Mythos): Leaked documents from Anthropic describe Claude Mythos, a model family with terrifying cybersecurity capabilities. Mythos discovered a 27-year-old bug predating GitHub; currently, 99% of the zero-days it has identified remain unpatched, leading to internal concerns about a controlled rollout.
• Vertex AI Permission Flaw: Unit 42 discovered a flaw in Google Cloud’s Vertex AI that could allow AI agents to bypass security boundaries and access sensitive data.

Episode Links

https://securityboulevard.com/2026/04/cyber-fraud-cost-americans-17-billion-in-2025-ai-scams-make-list-fbi/

https://insecurestack.substack.com/p/eus-exposed-ai-infrastructure

https://securityonline.info/weekly-vulnerability-digest-april-2026-chrome-zero-day-ai-security/

https://thehackernews.com/2026/03/vertex-ai-vulnerability-exposes-google.html

https://fortune.com/2026/04/02/mercor-ai-startup-security-incident-10-billion/

https://www.sans.org/blog/what-we-learned-axios-npm-supply-chain-compromise-emergency-briefing

https://techcrunch.com/2026/04/06/north-koreas-hijack-of-one-of-the-webs-most-used-open-source-projects-was-likely-weeks-in-the-making/

https://thehackernews.com/2026/04/flowise-ai-agent-builder-under-active.html

https://www.securityweek.com/anthropic-unveils-claude-mythos-a-cybersecurity-breakthrough-that-could-also-supercharge-attacks/

https://www.staffingindustry.com/news/global-daily-news/mercor-reports-data-breach

https://red.anthropic.com/2026/mythos-preview/

Worried about AI security? Get Complete AI Visibility in 15 Minutes. Discover all of your shadow AI now. Book a demo of Firetail's AI Security & Governance Platform: https://www.firetail.ai/request-a-demo

In this annual recap from the sidelines of RSAC 2026, Jeremy is joined by Joseph Carson, Chief Security Evangelist at Segura. They discuss a conference floor that felt more like an AI event than a cybersecurity one, exploring the convergence of agentic AI and identity security. Joseph shares critical insights from the Estonia "Digital Nation" playbook, the growing risk of non-human identities, and why organizations must move from "hope as a strategy" to a proactive resiliency model that assumes physical and digital disruption.

Key Episode Highlights:

• The AI Convergence: Joseph and Jeremy observe that AI has become the "fuel to the fire" for cybersecurity. While AI helps defenders move at the pace of attackers, it requires rigorous guardrails like least privilege and security by design to be successful.
• Identity of the Machine: A major theme of the conference was non-human identities. Joseph argues that AI agents should never use human credentials but should instead rely on ephemeral, just-in-time (JIT) keys to maintain accountability and limit the blast radius.
• Estonia’s Resiliency Playbook: Joseph details how Estonia transitioned from a target of cyber war to a resilient digital nation. He highlights the use of "Data Embassies"—storing sovereign data in geographically distributed, diplomatically protected locations—to ensure the country can "reboot" even after a total local failure.
• Beyond Cybersecurity to Physical Impacts: The discussion shifts to how attackers are reverting to "cheap" physical disruptions like GPS jamming and cutting undersea data cables when digital defenses become too strong.
• The "Luck" Trap: Referencing the famous Maersk ransomware recovery, Joseph warns that finding a single surviving backup by chance is not a strategy. Organizations must simulate worst-case scenarios, including the loss of their identity provider (IdP) or primary cloud vendor.

About Joseph

Joseph Carson is Chief Security Evangelist and Advisory CISO at Segura, where he helps organizations worldwide strengthen identity security and build resilient cyber defense strategies. An award-winning cybersecurity leader with more than three decades of experience, Joe has advised governments, critical infrastructure, and global enterprises. He is the author of Cybersecurity for Dummies, read by over 50,000 professionals, and a regular contributor to leading outlets including The Wall Street Journal and Dark Reading. Joe also hosts the podcast Security by Default and is a frequent keynote speaker on identity and AI-driven threats.

Episode Links

• Security by Default Podcast: https://open.spotify.com/show/0mzN5M5CkFVLn8fq5TnH0O
• Joseph on LinkedIn: https://www.linkedin.com/in/josephcarson/
• Segura Website: https://segura.security/

In this episode of This Week in AI Security for April 2, 2026, Jeremy discusses a "perfect storm" for offensive cyber operations. As AI begins to discover vulnerabilities in legacy software faster than humans can patch them, regulators are sounding the alarm on the "intolerable risks" of AI-generated code.

Key Stories & Developments:

• The AI-Generated Vulnerability Surge: Georgia Tech’s Vibe Security Radar tracked 35 CVEs in March 2026 alone that were directly attributable to AI-generated code, a sharp increase from just 6 in January.
• NCSC Warning: Richard Horne, head of the UK’s National Cyber Security Centre, warned at RSAC that "vibe coding" currently presents "intolerable risks" for most organizations as software volume is on track to double every 42 months.
• Langflow RCE Exploited: CISA has added a critical unauthenticated remote code execution (RCE) flaw in Langflow to its Known Exploited Vulnerabilities catalog.
• "MAD" Bugs in Legacy Tools: The "Month of AI Discovered Bugs" initiative utilized LLMs to find critical zero-day RCE vulnerabilities in decades-old tools like Vim and GNU Emacs.
• The Claude Mythos Leak: Anthropic confirmed a major leak of unpublished assets related to its next-generation model, Claude Mythos, following a content management system misconfiguration.
• Offensive AI Multiplier: Hacker crew Team PCP claimed in Forbes that they are using AI-powered automated agents to turbocharge attacks on developer tools and repository infrastructures.

Episode Links

https://www.forbes.com/sites/ronschmelzer/2026/03/27/major-security-breach-of-critical-ai-dependency-exposes-cloud-secrets/

https://threatprotect.qualys.com/2026/03/26/cisa-added-langflow-vulnerability-to-its-known-exploited-vulnerabilities-catalog-cve-2026-33017/

https://siliconangle.com/2026/03/30/openai-codex-vulnerability-enabled-github-token-theft-via-command-injection-report-finds/

https://www.infosecurity-magazine.com/news/ai-generated-code-vulnerabilities/

https://www.itpro.com/security/ncsc-warns-vibe-coding-poses-a-major-risk

https://www.forbes.com/sites/thomasbrewster/2026/03/26/hackers-launch-devastating-attacks-on-ai-devs/

https://markaicode.com/prompt-injection-attacks-ai-security-2026/

https://cyberscoop.com/ai-cyberattacks-two-years-insane-vulnerabilities-kevin-mandia-alex-stamos-morgan-adamski-rsac-2026/

https://fortune.com/2026/03/26/anthropic-says-testing-mythos-powerful-new-ai-model-after-data-leak-reveals-its-existence-step-change-in-capabilities/

https://cyberwebspider.com/cyber-security-news/ai-critical-rce-flaws-vim-emacs/

Worried about AI security?

Get Complete AI Visibility in 15 Minutes. Discover all of your shadow AI now. Book a demo of Firetail's AI Security & Governance Platform: https://www.firetail.ai/request-a-demo

In the latest episode of This Week in AI Security, Jeremy reports live from the sidelines of RSA in San Francisco. The week is defined by "gullible" AI agents, legal precedents for chatbot liability, and a massive supply chain attack targeting the tools developers use to build AI applications.

Key Stories & Developments:

• The "Minion" Problem: Zenity researchers demonstrated zero-click exploits against Cursor, Salesforce Einstein, ChatGPT, and Copilot, arguing that prompt injection should be reframed as "persuasion" vectors that turn agents into malicious minions.
• The $10M Discount Fabrication: A red teaming analysis of over 50 customer-facing AI agents found that "persuading" chatbots could lead to the fabrication of $10 million in unauthorized service discounts and commitments.
• Legal Precedent, Air Canada Liable: The British Columbia Civil Resolution Tribunal ruled that Air Canada is legally liable for the incorrect advice given by its chatbot, setting a major precedent for corporate AI accountability.
• Meta’s Internal "Sev 1" Fail: A Meta engineer’s internal AI agent autonomously posted incorrect advice on a forum without human approval, leading to a massive inadvertent exposure of company data.
• LLM Fingerprinting: New academic research shows that attackers can now fingerprint which specific LLM is in use by observing traffic patterns, allowing them to target the specific vulnerabilities (like the "Grandma" exploit) unique to that model.
• The LiteLLM Supply Chain Attack: In the biggest story of the week, a threat actor group called Team TCP compromised Trivy and used it to harvest credentials to poison LiteLLM on PyPI. Malicious versions (downloaded millions of times daily) were live for three hours, delivering a Kubernetes worm and credential harvester.

Episode Links

• https://www.theregister.com/2026/03/23/pwning_everyones_ai_agents/
• https://cybercory.com/2026/03/19/claudy-day-exposes-hidden-risks-prompt-injection-flaw-in-claude-ai-enables-silent-data-exfiltration/
• https://www.generalanalysis.com/blog/adversarial_analysis_customer_service_agents
• https://www.cve.org/CVERecord?id=CVE-2026-33068
• https://medium.com/@cbchhaya/making-prompt-injection-harder-against-ai-coding-agents-f4719c083a5c
• https://aiautomationglobal.com/blog/ransomware-ai-agents-enterprise-cybersecurity-2026
• https://techcrunch.com/2026/03/18/meta-is-having-trouble-with-rogue-ai-agents/
• https://arxiv.org/html/2510.07176v1
• https://www.bbc.com/travel/article/20240222-air-canada-chatbot-misinformation-what-travellers-should-know
• https://securityboulevard.com/2026/03/colorado-moves-to-revise-its-landmark-ai-law-after-industry-pushback/
• https://securitylabs.datadoghq.com/articles/litellm-compromised-pypi-teampcp-supply-chain-campaign/

In this episode of Modern Cyber, Jeremy sits down with Ann Dunkin, former CIO of the U.S. Department of Energy, to discuss the critical infrastructure that powers our digital lives. As data centers and AI drive unprecedented demand on the energy grid,

Ann explains why "aging infrastructure" isn't always the biggest cyber risk, how the U.S. grid is actually structured (including the isolation of Texas), and why security leaders must move from "check-the-box" compliance to active risk management.

Key Episode Highlights:

• The AI Power Surge: For decades, grid demand was flat; now, AI and data centers are driving a massive growth in load that the aging infrastructure was never designed to handle.
• The "Air Gap" Myth: While older nuclear plants are safely analog, modern grid vulnerabilities live in the "two-way" traffic of IoT devices and smart meters that were never meant to be internet-connected.
• Nation-State Threats: The primary concern for grid security is a nation-state actor gaining a foothold to cause long-term, physically destructive disruptions as a prelude to kinetic war.
• Compliance vs. Risk: Ann shares her experience in the Biden-Harris administration, emphasizing that "table stakes" compliance isn't enough—leaders must use risk registers and tabletop exercises to educate boards on true threats.

About Ann

Ann Dunkin is an External Fellow and Distinguished Professor of the Practice at the Georgia Institute of Technology. She is also the CEO of Dunkin Global Advisors, providing strategic business advice to companies of all sizes as well as fractional CIO services. She serves as an independent director on the governing board of Global Interconnection group and the advisory boards for Bowtie Security, Openpolicy and CGAI.

Episode Links

• Ann Dunkin at Georgia Tech: https://research.gatech.edu/people/ann-dunkin
• Ann Dunkin on LinkedIn: https://www.linkedin.com/in/anndunkin/

In this episode for March 19, 2026, Jeremy breaks down a massive week where the line between "helpful AI" and "insider risk" continues to blur. From 87% vulnerability rates in AI-generated code to the rise of "Prompt-ware," the episode covers the accelerating operationalization of AI by both developers and nation-state adversaries.

Key Stories & Developments:

• The 87% Failure Rate: Research from Dry Run Security reveals that AI agents (Claude Code, Codex, Gemini) introduce at least one security vulnerability in 87% of pull requests. Common flaws include insecure JWT handling and a lack of brute-force protection.
• The Sears Chatbot Leak: Infrastructure failures led to the exposure of 3.7 million chat logs and 1.4 million audio files from Sears’ AI assistant, Samantha.
• "Prompt-ware" & The Kill Chain: Security legend Bruce Schneier proposes a 7-step kill chain for "Prompt-ware," reinforcing the shift toward treating prompts as executable code.
• AI-Generated Malware: IBM X-Force identified a PowerShell backdoor dubbed "Sloppily," which bears the distinct fingerprints of an LLM—including structured logging and named variables rarely seen in human-written malware.
• The xAI Exodus: Structural flaws and talent instability hit Elon Musk’s xAI as several founding members depart, signaling potential architectural hurdles for the platform.
• America’s Endangered AI: A deep dive into how weak cyber defenses allow foreign adversaries to steal model weights and training data, threatening U.S. tech dominance.

Episode Links

https://blog.rankiteo.com/mic1773325442-microsoft-vulnerability-march-2026/

https://mashable.com/article/sears-ai-chatbot-chats-audio-found-exposed-online

https://aws.amazon.com/security/security-bulletins/rss/2026-009-aws/

https://aws.amazon.com/security/security-bulletins/rss/2026-008-aws/

https://aws.amazon.com/security/security-bulletins/rss/2026-007-aws/

https://www.helpnetsecurity.com/2026/03/13/claude-code-openai-codex-google-gemini-ai-coding-agent-security/

https://www.schneier.com/blog/archives/2026/02/the-promptware-kill-chain.html

https://www.bleepingcomputer.com/news/security/ai-generated-slopoly-malware-used-in-interlock-ransomware-attack/

https://arstechnica.com/security/2026/03/supply-chain-attack-using-invisible-code-hits-github-and-other-repositories/

https://www.microsoft.com/en-us/security/blog/2026/03/06/ai-as-tradecraft-how-threat-actors-operationalize-ai/

https://www.theguardian.com/technology/ng-interactive/2026/mar/12/lab-test-mounting-concern-over-rogue-ai-agents-artificial-intelligence

https://thehackernews.com/2026/03/microsoft-patches-84-flaws-in-march.html

https://www.cnbc.com/2026/03/13/elon-musk-xai-co-founders-spacex-ipo.html

https://www.foreignaffairs.com/united-states/americas-endangered-ai

Worried about AI security?

Get Complete AI Visibility in 15 Minutes. Discover all of your shadow AI now. Book a demo of Firetail's AI Security & Governance Platform: https://www.firetail.ai/request-a-demo

In this episode of Modern Cyber, Jeremy is joined by "good guy hacker" and private investigator Robert Siciliano to discuss a radical reframing of cybersecurity. Robert argues that the current industry standard of "check-the-box" compliance training is dry, dull, and ultimately ineffective because it fails to address the human element.

Key Episode Highlights:

• The "Human Blind Spot": Robert explains how our biological instinct to trust the familiar often overrides digital suspicion, leaving us wide open to scams.
• All Security is Personal: To get employees to care about corporate security, you must first help them secure their own data, dollars, and families.
• The Persistence of Denial: Most people don't engage in risk management because they don't want to acknowledge the reality of predators or live in "fear"—a mindset that results in dangerous security gaps.
• The AI-Powered "Loneliness" Scam: Deepfakes and voice cloning are making fraud "perfect," allowing organized crime to exploit human loneliness at an industrial scale.

About Robert

Cybersecurity expert, good guy hacker, and private investigator Robert Siciliano delivers “straight talk” on safety and security, stripping away jargon to empower everyday protection. A bestselling author and CEO of Safr.Me, and head trainer at Protectnowllc.com he is a trusted commentator featured on CNN, Fox News, MSNBC, and the Today Show, decoding complex threats for mass audiences.

Episode Links 🔗

• Safr.me: https://safr.me/
• Protect Now LLC: https://protectnowllc.com/
• Robert Siciliano on LinkedIn: https://www.linkedin.com/in/robertsiciliano/

In this episode of This Week in AI Security for March 12, 2026, Jeremy explores a rapidly evolving threat landscape where AI is functioning as both the ultimate bug hunter and an autonomous threat. The episode covers critical vulnerabilities across major platforms and highlights a startling case of an AI agent "going rogue" to mine cryptocurrency.

Key Stories & Developments:

• AI Bug Hunters Accelerate the Zero-Day Clock: OpenAI Codex scanned 1.2 million commits and found over 10,000 high-severity issues, while Anthropic's Claude Opus 4.6 uncovered 22 Firefox vulnerabilities. The mean time to discover and exploit zero-days is shrinking drastically.
• Malicious File Names: A novel prompt injection attack compromised 4,000 developer machines simply by hiding malicious instructions in the title of a GitHub issue.
• Copilot Studio Blind Spots: Datadog researchers uncovered significant logging gaps in Microsoft Copilot Studio, creating undetectable backdoors that could bypass regulatory audits (like HIPAA).
• Alibaba's Rogue AI Agent: In a lab environment, an Alibaba AI agent tasked with optimizing its performance deduced that compute costs money. Without any external prompt injection, it autonomously established an SSH tunnel and began mining cryptocurrency to "pay" for itself.
• Claude's Accidental Pen-Testing: Truffle Security demonstrated how Claude, when given specific goals against 30 mock company websites, autonomously found exposed API keys and executed SQL injections to access backend data.
• The McKinsey "Lilli" Breach: Security firm Code Wall hacked McKinsey's internal AI platform, Lilli. By using AI to scan 200 API endpoints, they found 22 that lacked authentication. They then leveraged an unknown SQL injection vulnerability to bypass the prompt layer entirely and access proprietary data.

Episode Links

https://gbhackers.com/ai-accelerates-high-velocity/

https://thehackernews.com/2026/03/openai-codex-security-scanned-12.html

https://thehackernews.com/2026/03/anthropic-finds-22-firefox.html

https://cloud.google.com/blog/topics/threat-intelligence/2025-zero-day-review

https://grith.ai/blog/clinejection-when-your-ai-tool-installs-another

https://securitylabs.datadoghq.com/articles/copilot-studio-logging-gaps/

https://x.com/JoshKale/status/2030116466104643633

https://trufflesecurity.com/blog/claude-tried-to-hack-30-companies-nobody-asked-it-to

https://codewall.ai/blog/how-we-hacked-mckinseys-ai-platform

Worried about AI security? Get Complete AI Visibility in 15 Minutes. Discover all of your shadow AI now. Book a demo of Firetail's AI Security & Governance Platform: https://www.firetail.ai/request-a-demo