chore: update feed state [2026-05-23]

Author: github-actions[bot]

logs/urls.txt

@@ -432,3 +432,15 @@
 [2026-05-22] https://www.datadoghq.com/blog/cloud-cost-management-budget-forecasting/
 [2026-05-22] https://www.wiz.io/blog/claude-wiz-integration
 [2026-05-22] https://secret.club/2026/05/21/striga.html
+[2026-05-23] https://outpost24.com/blog/combining-easm-and-pentesting/
+[2026-05-23] https://shellsharks.com/scrolls/scroll/2026-05-22
+[2026-05-23] https://blog.nviso.eu/2026/05/22/securing-ai-systems-without-overconfidence-or-fear-part-2-attack-surfaces-and-the-checkpoint-flow/
+[2026-05-23] https://blog.fox-it.com/2026/05/22/remotepe-the-lazarus-rat-that-lives-in-memory/
+[2026-05-23] https://blog.trailofbits.com/2026/05/22/we-hardened-zizmors-github-actions-static-analyzer/
+[2026-05-23] https://unit42.paloaltonetworks.com/tracking-iran-apt-screening-serpens/
+[2026-05-23] https://unit42.paloaltonetworks.com/roadtools-cloud-attacks/
+[2026-05-23] https://www.sentinelone.com/blog/the-good-the-bad-and-the-ugly-in-cybersecurity-week-21-7/
+[2026-05-23] https://www.sentinelone.com/blog/sentinels-league-live-rankings/
+[2026-05-23] https://www.datadoghq.com/blog/securing-ai-agents-guardrail-placement/
+[2026-05-23] https://www.datadoghq.com/blog/improve-api-authentication-detection-with-datadog/
+[2026-05-23] https://www.datadoghq.com/blog/how-to-measure-developer-experience-in-the-ai-era/

site/_posts/2026-05-22-paved-with-intent-roadtools-and-nation-state-tactics-in-the.md

@@ -0,0 +1,12 @@
+---
+layout: post
+title: "Paved With Intent: ROADtools and Nation-State Tactics in the Cloud"
+date: 2026-05-22 10:00:24 +0300
+categories: [RSS]
+tags: [cloud, entra-id, oauth, token-abuse, detection]
+toc: true
+---
+
+Unit 42 analyzes how attackers operationalize ROADtools against Microsoft Entra ID, focusing on two core modules: ROADrecon for tenant enumeration and roadtx for token acquisition, exchange, and device registration. The tooling works through legitimate Microsoft APIs and supports OAuth/OIDC flows such as device code, refresh token reuse, and on-behalf-of exchanges, which lets adversaries replay or manipulate tokens, persist access, and in some cases bypass MFA while blending into normal cloud traffic with customizable user-agent strings. The article highlights how ROADrecon maps users, groups, roles, devices, service principals, and applications into a local SQLite-backed graph of privileged relationships, making it useful for discovery and escalation planning even as the ecosystem shifts from Azure AD Graph to fragmented Microsoft Graph forks. This is high-signal cloud identity tradecraft because it ties publicly available tooling to recent nation-state intrusions and pairs the offensive mechanics with concrete hunting opportunities for detecting ROADtools-style Entra ID activity.
+
+[Read original article](https://unit42.paloaltonetworks.com/roadtools-cloud-attacks/){: .btn .btn-primary }

site/_posts/2026-05-22-remotepe-the-lazarus-rat-that-lives-in-memory.md

@@ -0,0 +1,12 @@
+---
+layout: post
+title: "RemotePE: The Lazarus RAT that lives in memory"
+date: 2026-05-22 14:55:58 +0300
+categories: [RSS]
+tags: [lazarus, dpapi, rat, malware, edr-evasion]
+toc: true
+---
+
+Fox-IT analyzes a Lazarus-linked three-stage Windows implant chain made up of DPAPILoader, RemotePELoader, and RemotePE, where the final RAT is executed entirely in memory and never written to disk. The first-stage DLL masquerades as a legitimate Windows service component (for example `C:\Windows\System32\Iassvc.dll` under the fake "Internet Authentication Service"), searches `C:\ProgramData\Microsoft\Windows\DeviceMetadataStore\en-US*.*` for non-`MSCF` files larger than 50 KiB, then decrypts the payload with Windows DPAPI and an additional XOR `0x8D` layer before reflective loading via libpeconv. That DPAPI-based environmental keying binds payloads to the victim machine, frustrates VirusTotal and static analysis, and produces per-victim ciphertext that defeats simple hash-based detection. The recovered second stage polls C2 for the in-memory RemotePE payload and applies syscall-based evasion using the TartarusGate variant of HellsGate, indicating a low-footprint, long-dwell intrusion set tailored for stealthy financial and cryptocurrency targeting.
+
+[Read original article](https://blog.fox-it.com/2026/05/22/remotepe-the-lazarus-rat-that-lives-in-memory/){: .btn .btn-primary }

site/_posts/2026-05-22-securing-ai-agents-why-guardrail-placement-is-a-key-design-d.md

@@ -0,0 +1,12 @@
+---
+layout: post
+title: "Securing AI agents: Why guardrail placement is a key design decision"
+date: 2026-05-22 00:00:00 +0300
+categories: [RSS]
+tags: [ai-agents, prompt-injection, llm-security, aws, guardrails]
+toc: true
+---
+
+This article analyzes how guardrail placement inside an AI agent’s orchestration loop changes the ability to stop indirect prompt injection attacks, using a demo where a benign request to summarize a GitHub issue causes the agent to read attacker-controlled issue text and then invoke a secret-reading tool. The Bedrock-specific example shows that developer-added checks often end up in the Action Group Lambda and must call the ApplyGuardrail API on tool output (`source="OUTPUT"`), because that Lambda only receives the current tool invocation’s parameters rather than the full prompt, memory, and conversation history. That visibility gap means the guardrail can inspect or filter individual tool responses but cannot reason over the entire agent state at earlier decision points where the model decides to call `GetLocalSecret` in the first place. The post contrasts this with self-orchestrated agents that can insert evaluations at multiple hook points around prompt construction, tool execution, and response generation, making it easier to detect prompt-injection-driven control-flow changes instead of just sanitizing downstream content.
+
+[Read original article](https://www.datadoghq.com/blog/securing-ai-agents-guardrail-placement/){: .btn .btn-primary }

site/_posts/2026-05-22-the-good-the-bad-and-the-ugly-in-cybersecurity-week-21.md

@@ -0,0 +1,12 @@
+---
+layout: post
+title: "The Good, the Bad and the Ugly in Cybersecurity – Week 21"
+date: 2026-05-22 15:08:13 +0300
+categories: [RSS]
+tags: [macos, infostealer, windows, privilege-escalation, defender]
+toc: true
+---
+
+This weekly roundup combines several security items, but the most technically useful sections cover a new SHub "Reaper" macOS stealer variant and two actively exploited Microsoft Defender zero-days. The macOS malware uses fake WeChat and Miro installers on typosquatted domains, anti-analysis checks, and the `applescript://` URL scheme to launch Script Editor, pad malicious code with ASCII art, prompt for a user password to access Keychain data, steal browser and iCloud credentials, and exfiltrate files in 70MB ZIP chunks; it also hijacks crypto apps by replacing `app.asar`, clearing quarantine attributes, and applying ad hoc signing to bypass Gatekeeper. The Windows section summarizes CVE-2026-41091, a link-following flaw in the Microsoft Malware Protection Engine that can yield SYSTEM privileges, and CVE-2026-45498, a Defender Antimalware Platform bug that can force denial-of-service conditions on unpatched systems. The law-enforcement coverage on Operation Ramz, an infostealer operator case, and Europol's takedown of the "First VPN" service adds context on current cybercrime disruption, but the article is primarily a curated digest rather than a deep original technical write-up.
+
+[Read original article](https://www.sentinelone.com/blog/the-good-the-bad-and-the-ugly-in-cybersecurity-week-21-7/){: .btn .btn-primary }

site/_posts/2026-05-22-tracking-iranian-apt-screening-serpens-2026-espionage-campai.md

@@ -0,0 +1,12 @@
+---
+layout: post
+title: "Tracking Iranian APT Screening Serpens’ 2026 Espionage Campaigns"
+date: 2026-05-22 13:00:42 +0300
+categories: [RSS]
+tags: [apt, iran, rat, phishing, .net]
+toc: true
+---
+
+Unit 42 tracks an Iran-linked Screening Serpens espionage campaign that used tailored job-recruitment phishing lures against aerospace, defense, telecom, and technology targets in the U.S., Israel, the UAE, and other Middle Eastern organizations during February-April 2026. The report identifies six new RAT variants across two malware families, MiniUpdate and MiniJunk V2, delivered through infection chains that rely on DLL sideloading and per-target command-and-control domain sets, many hosted on Azure, to improve operational resilience and reduce cross-contamination between campaigns. A key technical evolution is the use of AppDomainManager hijacking in .NET applications, where a legitimate configuration file is abused during application initialization to disable built-in security controls before the malware loads. The research is useful because it documents both the actor’s malware-development cadence and a concrete tradecraft upgrade that defenders can hunt for in phishing-driven .NET execution chains.
+
+[Read original article](https://unit42.paloaltonetworks.com/tracking-iran-apt-screening-serpens/){: .btn .btn-primary }

site/_posts/2026-05-22-we-hardened-zizmors-github-actions-static-analyzer.md

@@ -0,0 +1,12 @@
+---
+layout: post
+title: "We hardened zizmor's GitHub Actions static analyzer"
+date: 2026-05-22 11:00:00 +0300
+categories: [RSS]
+tags: [ci-cd, yaml, static-analysis, supply-chain]
+toc: true
+---
+
+Trail of Bits describes hardening `zizmor`, a static analyzer for GitHub Actions workflows, after GitHub's addition of YAML anchors exposed parsing and analysis gaps in a tool meant to catch CI/CD misconfigurations like `pull_request_target` abuse. Using a corpus of 41,253 workflow files from 6,612 high-value open-source repositories, they found and fixed four anchor-related bugs: aliases in sequences being flattened incorrectly, anchor prefixes leaking into resolved values, duplicate anchors causing panics, and the `template-injection` audit crashing when `run:` used an aliased scalar. The work also uncovered deserialization edge cases in otherwise valid workflows such as `if: 0`, `timeout-minutes: 0.5`, and `secrets: inherit`, plus expression-evaluator mismatches that were corrected against GitHub's Known Answer Tests. This is practically relevant because `zizmor` is used to detect workflow security issues before deployment, and robustness around real-world YAML and expression handling directly affects defenders' ability to catch supply-chain-relevant GitHub Actions misconfigurations.
+
+[Read original article](https://blog.trailofbits.com/2026/05/22/we-hardened-zizmors-github-actions-static-analyzer/){: .btn .btn-primary }

state/processed_urls.json

@@ -1646,7 +1646,19 @@
     "https://securitylabs.datadoghq.com/articles/unpatchable-kubernetes-vulnerabilities-cve-2021-25740/": "2026-05-22",
     "https://www.datadoghq.com/blog/cloud-cost-management-budget-forecasting/": "2026-05-22",
     "https://www.wiz.io/blog/claude-wiz-integration": "2026-05-22",
-    "https://secret.club/2026/05/21/striga.html": "2026-05-22"
+    "https://secret.club/2026/05/21/striga.html": "2026-05-22",
+    "https://outpost24.com/blog/combining-easm-and-pentesting/": "2026-05-23",
+    "https://shellsharks.com/scrolls/scroll/2026-05-22": "2026-05-23",
+    "https://blog.nviso.eu/2026/05/22/securing-ai-systems-without-overconfidence-or-fear-part-2-attack-surfaces-and-the-checkpoint-flow/": "2026-05-23",
+    "https://blog.fox-it.com/2026/05/22/remotepe-the-lazarus-rat-that-lives-in-memory/": "2026-05-23",
+    "https://blog.trailofbits.com/2026/05/22/we-hardened-zizmors-github-actions-static-analyzer/": "2026-05-23",
+    "https://unit42.paloaltonetworks.com/tracking-iran-apt-screening-serpens/": "2026-05-23",
+    "https://unit42.paloaltonetworks.com/roadtools-cloud-attacks/": "2026-05-23",
+    "https://www.sentinelone.com/blog/the-good-the-bad-and-the-ugly-in-cybersecurity-week-21-7/": "2026-05-23",
+    "https://www.sentinelone.com/blog/sentinels-league-live-rankings/": "2026-05-23",
+    "https://www.datadoghq.com/blog/securing-ai-agents-guardrail-placement/": "2026-05-23",
+    "https://www.datadoghq.com/blog/improve-api-authentication-detection-with-datadog/": "2026-05-23",
+    "https://www.datadoghq.com/blog/how-to-measure-developer-experience-in-the-ai-era/": "2026-05-23"
   },
-  "last_updated": "2026-05-22T03:16:27.744088+00:00"
+  "last_updated": "2026-05-23T03:13:52.802400+00:00"
 }