Jeremy Kirk
@jkirk.bsky.social
Okta Threat Intelligence. Personal account. Interests: Cyber threat intelligence, OSINT, data breaches, AI. Formerly intel analysis @ Intel 471. jeremykirk.com
Behind adverts for suspiciously cheap AI subscriptions for Anthropic's Claude, OpenAI's ChatGPT, Cursor, Gemini, etc., is an ecosystem of fakery and fraud. Okta Threat Intelligence profiled the account signup fraud TTPs and took disruptive action: www.okta.com/en-gb/blog/t...
Free tokens for sale: How fake signups drive AI fraud
Bad actors are registering fake accounts to resell discounted and trial AI services from Anthropic, Google and Amazon for profit. Here's the lowdown.
okta.com
Okta Threat Intelligence obtained from a sensitive source a "vishing" kit used to hijack enterprise accounts. This is a rare peek inside the software that cybercriminals have developed for themselves in order to scale identity-based attacks: www.okta.com/blog/threat-...
Behind the scenes of a vishing operation
okta.com
OpenAI's models discovered an 0-day and used other vulnerabilities to break out of a sandbox in order to cheat on an AI benchmarking test hosted on Hugging Face. 😲 openai.com/index/huggin...
OpenAI and Hugging Face partner to address security incident during model evaluation
OpenAI and Hugging Face share early findings from a security incident during AI model evaluation, highlighting advanced cyber capabilities and lessons for defenders.
openai.com
A LLM cliché highlighter app by @simonwillison.net. 🤣 tools.simonwillison.net/llm-cliche-h...
LLM cliché highlighter
tools.simonwillison.net
Hugging Face: "The practical lesson for defenders: have a capable model you can run on your own infrastructure vetted and ready before an incident, both to avoid guardrail lockout and to keep attacker data and credentials from leaving your environment." huggingface.co/blog/securit...
Security incident disclosure — July 2026
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
Court Listener has an MCP server for Claude and OpenAI: wiki.free.law/c/courtliste...
Model Context Protocol (MCP) Server for Agentic Access
Use an MCP to connect AI tools and agents like Claude and ChatGPT to CourtListener's full functionality, enabling complex legal monitoring, research, and analysis.
wiki.free.law
"Invariably an agentic codebase will end up larger and more overwrought than anything built by human hand. This is technical debt on an unprecedented scale, accrued at machine speed." by @wesmckinney.com wesmckinney.com/blog/mythica...
The Mythical Agent-Month – Wes McKinney
wesmckinney.com
Significant disruption of the SocGholish, Amadey and StealC loaders/infostealers as part of the ongoing Operation Endgame law enf operation. These malware fuel initial access and cred theft: www.europol.europa.eu/media-press/...
Global cyber strike disrupts SocGholish, Amadey, and StealC malware networks – Coordinated actions take down criminal infrastructure; over EUR 41 million in criminal crypto assets seized | Europol
Europol together with partners from across the globe today announces a landmark blow to cybercriminal networks as part of Operation Endgame, a sweeping international operation targeting the criminal i...
europol.europa.eu
A dark economy will emerge around serving open-weight models such as GLM-5.2 via API to threat actors. Defenders need access to frontier models to get ahead in adopting AI for better cybersecurity, argues Joshua Saxe, formerly of AI testing at Meta. joshuasaxe181906.substack.com/p/glm-52-not...
GLM-5.2, not Mythos, is the real security emergency
Until last week, attackers faced a dilemma in using frontier models: even if they could manage the cat-and-mouse game of setting up fake accounts to retain API access to frontier model providers, and even if they could induce models to help them hack via creative prompting, their usage was logged, so if discovered after the fact, their tactics, techniques, procedures, goals, and targets would be exposed to defenders.
joshuasaxe181906.substack.com
Warning from Five Eyes about the vulnerability storm coming due to AI-driven bug hunting: www.cyber.gov.au/about-us/vie...
cyber.gov.au
GitHub has made several changes around actions/checkout to reduce misuse of pull_request_target, or the infamous pwn request: github.blog/changelog/20...
Safer pull_request_target defaults for GitHub Actions checkout - GitHub Changelog
The pull_request_target event is one of the most commonly misused triggers in GitHub Actions, leading to vulnerabilities in workflows. Workflows triggered by pull_request_target run with the base repo...
github.blog
Yet another incident that should motivate SaaS app developers to support DPoP for OAuth, which cryptographically ties OAuth tokens to the client they were issued to. klue.com/blog/an-upda...
An Update on the Recent Klue Security Incident - Klue
An update from Klue's CEO on a recent security incident.
klue.com
Similar to OpenAI's recent post about safe coding with AI, Anthropic has published how it keeps the coding genie in a bottle: www.anthropic.com/engineering/...
How we contain Claude across products
Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.
anthropic.com
Went to a venue the other night. Checked in a nice, wide Bunnings umbrella at the door. Usher put my name on it. Someone nicked it anyway. Very cross. #australia
GitHub's supply chain incident is linked to TanStack's. A Nx dev's Github creds were compromised in that incident, which then led to the publishing of the malicious VS Code extension: github.com/nrwl/nx-cons...
Compromised Nx Console version 18.95.0
Update (May 19 13:37 UTC): Updated the timeline of the compromised VSCode extension. Added information about OpenVSX. Update (May 20 23:46 UTC): Updated details on scope of impact Update (May 21 ...
github.com
Come work with us at Okta ! We're looking for a new member of Okta's Threat Intelligence team. This role is one in which if you have a good idea that fits our mission, you can run with it. Plus, we're nice people. 😀 North Korean IT workers need not apply. www.linkedin.com/jobs/view/44...
Okta hiring Director, Okta Threat Intelligence in Bellevue, WA | LinkedIn
Posted 1:35:03 AM. Secure Every Identity, from AI to HumanIdentity is the key to unlocking the potential of AI. Okta…See this and similar jobs on LinkedIn.
linkedin.com
Device-code phishing is surging. One cybercriminal tool can target virtually every Microsoft user. Okta Threat Intelligence digs deep into the OAuth CLI device flow and how to minimize your attack surface. www.okta.com/blog/threat-...
Device code phishing: it’s phishing with dynamite
Okta Threat Intelligence studies recent abuse of the device code authorization flow in service provider ecosystems, and provides links to detection and remediation strategies for protecting clients cr...
okta.com
Good post on OpenAI's security controls around Codex: Sandboxing, approvals, limited network access, OAuth for CLI/MCP, dangerous shell commands blocked, OpenTelemetry log exports for prompts, tools, MCP usage, etc. centralized in SIEM. openai.com/index/runnin...
Running Codex safely at OpenAI
How OpenAI runs Codex securely with sandboxing, approvals, network policies, and agent-native telemetry to support safe and compliant coding agent adoption.
openai.com
Device code phishing is exploding, and AiTM actors are getting in on it. We found ODx phishing-as-a-service providing device code capabilities in addition to their AiTM offerings. ODx is one of the most popular AiTM kits currently. It's also tracked as Storm-1167 and FlowerStorm.
Corgi, an insurance company, is now covering AI for situations such as when your "AI agents go rogue," an agent that leaks customer data or copyright claims. 😮 www.linkedin.com/posts/nico-l...
Today, we're excited to launch AI Coverage (insurance for when your AI messes up). Insurance was built for risks that have existed for decades. AI is creating a new category very quickly. Think…...
Today, we're excited to launch AI Coverage (insurance for when your AI messes up). Insurance was built for risks that have existed for decades. AI is creating a new category very quickly. Think abou...
linkedin.com
Claude AI workloads can now use short-lived OIDC tokens issued by your IdP of choice (hopefully Okta!) rather than using static API keys, which is better for security. platform.claude.com/docs/en/buil...
Workload Identity Federation
Authenticate workloads to the Claude API with short-lived identity tokens from your own identity provider instead of long-lived static API keys.
platform.claude.com
Ran a two-hour Spotlight OSINT workflow using Deepseek-V4-Pro via its API with Claude as a harness. Quite fast. Cost US$.33 versus probably at least $5 on extra usage w/ Sonnet. No charge for using Claude as an orchestration layer. Worked very well. spotlight.buriedsignals.com
Spotlight — OSINT investigation system for journalists
Turn your agent into an investigative system. Methodology, investigator and fact-checker loops, human review gates, local-first workflows.
spotlight.buriedsignals.com
Over several weeks, we at Okta tested OpenClaw with various AI models to see how agents handle API keys, OAuth tokens, credentials and other secrets. The short of it is agents can't be trusted, and it's easy to talk agents into skirting their guardrails. More here: www.okta.com/newsroom/art...
Phishing the agent: Why AI guardrails aren’t enough
AI agents are being handed the "keys to the kingdom," but research shows they can't be trusted to hold them. Learn how agent guardrails can fail.
okta.com
Good post by @andrewnez.bsky.social about the supply-chain issues around GitHub Actions: nesbitt.io/2026/04/28/g...
GitHub Actions is the weakest link
Anne Robinson would like a word with .github/workflows
nesbitt.io
Tested a powerful OSINT tool last night by Buried Signals called Spotlight. Qwen 3.6 was being finicky on Ollama so ran it with Sonnet. Spotlight is an amazing investigative tool -- finds, confirms, reconfirms, dispels -- all incredibly fast. spotlight.buriedsignals.com
Spotlight — OSINT investigation system for journalists
Turn your agent into an investigative system. Methodology, investigator and fact-checker loops, human review gates, local-first workflows.
spotlight.buriedsignals.com
This is a well-reasoned piece for how news/content needs to dramatically change in the agentic AI age if it's to bring in any revenue. x402 is nice in theory but in this use case hinges on micropayments (with crypto 😬) for content, which never worked. fdaudens.substack.com/p/the-click-...
Sell to the Agent
A monetization playbook for the post-browser era
fdaudens.substack.com
When open-source AI models reach US frontier model capabilities, “every 19-year-old in St. Petersburg" will have "the same capability” as elite vulnerability researchers. cyberscoop.com/ai-cyberatta...
Security leaders say the next two years are going to be 'insane'
Top security experts warn AI is discovering vulnerabilities exponentially faster than defenders can respond, creating a "perfect storm" for attackers over the next two years.
cyberscoop.com
As if the information space isn't confusing enough these days... www.theverge.com/tech/896490/...
Google Search is now using AI to replace headlines
Let us know if you see more.
theverge.com
Moxie Marlinspike's Confer project — which aims to bring end-to-end encryption to protect the privacy of AI chats that are now usually being consumed by AI companies for training — will work to integrate it with Meta AI. #infosec confer.to/blog/2026/03...
Confer is bringing foundational AI privacy to Meta
I started building Confer because I saw how amazing LLMs are, and as a result, how much of our data is flowing through them. Already, AI chat apps have become some of the largest centralized data lake...
confer.to
