AI Security
34 posts tagged #ai-security
34 posts on AI Security, spread across AI Agent Architecture & Frameworks (10), AI Security & Safety (6), Local LLM Hardware (4) — most read: “DeepMind AlphaProof Code: What It Really Proved on Signal” and “AI Coding Agent Full IDE Access: 5 Ways It Breaks”.
AI and Machine Learning DeepMind AlphaProof Code: What It Really Proved on Signal
DeepMind's AlphaProof Code used Gemini-powered agents and 6M lines of Lean to verify Signal's protocol. Here's what was actually proven — and the 'zero bugs' claim it wasn't.
Developer Tools AI Coding Agent Full IDE Access: 5 Ways It Breaks
Giving an AI coding agent full IDE access sounds efficient until it expands scope, cascades refactors, and leaks data. Here are the five failure modes and a permission-tiering model that actually holds.
Developer Tools Docker Alternative for AI Agent Sandboxes: 3 Compared
Docker alone doesn't scope an AI coding agent's files, network, or tool calls. Here's how NVIDIA OpenShell, Anthropic sandbox-runtime, and Docker Sandboxes compare — and the factual error circulating about OpenShell's architecture.
AI and Machine Learning Uncensored Local LLMs: Abliteration Explained
Cut through the viral "forbidden AI" hype. Here's how abliteration strips refusals, whether running an uncensored local LLM is legal, and how to run one with Ollama.
Cybersecurity Local LLM Privacy Risk: What Your Setup Leaks
Two viral videos claim your local AI isn't private. I ran packet captures on Ollama to verify the three real leak vectors: open APIs, 0.0.0.0 binding, and client telemetry.
Cybersecurity DeepSeek Banned From US Federal Agencies: The DHS Order
A DHS order made public on Oct 19, 2026 bans DeepSeek from every federal civilian agency. Here's what the order actually says, the open-weights gray zone, and how to audit your own stack for foreign AI model dependencies.
Technology Anthropic Warns AI Models Could Soon Self-Replicate
Anthropic warns AI models could soon copy themselves onto small devices without a data center. Here's what 'self-replication' actually tests for, and how worried you should be.
Technology Ban Artificial Superintelligence Act: 20 Years in Prison
Sanders and Casar's Ban Artificial Superintelligence Act would jail AI executives for up to 20 years and pause frontier development. Here's what it actually says.
Cybersecurity How to Detect AI-Generated Bug Bounty Reports
A maintainer-first playbook to score report quality, spot LLM-written spam, add proof-of-work friction, rate-limit intake on GitHub, and reproduce safely in a sandbox.
Cybersecurity How to Ship EU AI Act Compliance for AI Agents
A shippable, vendor-neutral evidence pack for AI agents: run logs, tool approvals, guardrail configs, retention/redaction, and an audit export you can produce in days.
Developer Tools MCP vs Function Calling in Agents: When to Say No
MCP backlash is real. Here’s my decision framework for when to use MCP vs function calling in agents, and when a bespoke tool API beats both on ops, security, and debugging.
Cybersecurity How to Add Kill Switches and Spend Limits to AI Agents
A software-only watchdog pattern for AI agents: a single enforcement point for kill switches, per-tool budgets, approval gates, and tamper-evident audit logs.
Cybersecurity MCP OAuth Security: Impersonation, aud Mismatch, Replay
A threat model for MCP tool servers using OAuth: how tool impersonation and audience mismatch happen, how tokens get replayed, and what to validate and log.
Cybersecurity How to Stop Repo Prompt Injection in Coding Agents
Repo-level prompt injection turns “clone and ask the agent” into a supply-chain compromise. Here’s the threat model, a safe demo, and practical mitigations.
Cybersecurity MCP Server Security Best Practices: Checklist + CI Linter
A practical MCP server security test plan: authn/authz models, tool allowlists, prompt-injection regression tests, rate limits, audit logs, and a CI-friendly permissions linter.
Cybersecurity RatHat Android Malware “AI-Powered” Claims: What to Detect
RatHat is getting labeled “AI-powered,” but the real story is the Android stealer kill chain: accessibility abuse, permission traps, and noisy C2 loops you can actually detect.
Cybersecurity How to Secure Local LLM Inference: Sandbox + Egress
A practical blueprint for secure local LLM inference: sandbox inference hard, default-deny outbound network, stage allowlisted downloads, scan artifacts, and isolate tools.
AI and Machine Learning AI Agent Swarm Patterns 2026: Topologies and Failure Modes
AI agent swarm coordination patterns decide whether your multi-agent system converges or spirals. Here are the topologies, stop conditions, budget caps, and failure modes most teams don’t test in 2026.
Cybersecurity How to Implement OWASP Agentic Top 10 Controls
A control-by-control guide to OWASP agentic top 10 controls: where to gate tool calls (MCP/client/server), what to log, what to block, and what belongs in CI vs runtime.
Developer Tools AI Code Review Checklist: How to Review AI-Generated Code
AI makes code cheap. Verification is the new tax. Here’s a practical workflow, checklist, and ‘delete & redo’ rubric to keep PRs fast and safe in 2026.
![system logs terminal laptop screen code — illustration for article on LLM Data Leakage Playbook [2026]:](https://img.kunalganglani.com/images/vzekdneq/production/6d5e5f730983ecdcec86a90f783fd63f70b2bd9b-1200x675.webp?w=600&q=75&auto=format)
Cybersecurity LLM Data Leakage Playbook: Logging, Retention, Redaction
A practitioner playbook for preventing data leakage in LLM apps by hardening logging, retention, and redaction across the entire prompt→tools→model→observability path, with audit-ready evidence you can hand to compliance.
Cybersecurity AI Security Playbook for Leaders: 10 Controls That Ship
A practical AI security leader playbook you can implement this quarter: inventory, approval gates, agent threat modeling, OWASP LLM Top 10 controls, vendor review, and incident response.
AI and Machine Learning RAG Data Leakage Test Suite: CI Red-Team Setup
Build an automated red-team suite for RAG apps: canary tokens, regex + similarity detectors, multi-step prompt-injection attacks, and a CI risk score that blocks risky merges.
AI and Machine Learning AI Agent Sandbox on a Linux VM: Safe Tool Use, No K8s
If your coding agent can run `git`, `pip`, or a shell, it deserves its own disposable Linux VM. Default-deny egress, snapshot rollback, scoped secrets, and per-run audit bundles. No Kubernetes required.
![claude code terminal laptop screen — illustration for article on Claude Code Security [2026]: Risks, Safe](https://img.kunalganglani.com/images/vzekdneq/production/ae489a760c8efd343421ac3a5e4b1f9b048acc2c-1200x675.webp?w=600&q=75&auto=format)
Technology Claude Code Security: Risks, Safe Setup, Team Policy
Claude Code is safe only if you treat it like a junior engineer with terminal access. Here’s the 2026 playbook: permissions, sandboxing, egress controls, MCP allowlists, retention settings, and incident response.
Cybersecurity AI Agent Memory Exfiltration: Kill Chain + 5-Step Hardening
A proof-of-concept showed Claude's memory could be silently exfiltrated to an attacker's server with no user warning. Here's the full kill chain, which memory architectures are vulnerable, and a 5-step hardening checklist grounded in OWASP LLM Top 10 2025.
![AI Voice Detector: Detect AI Audio & Speech [2026]](https://img.kunalganglani.com/images/vzekdneq/production/a491086937bfeaf51b5d07e647cc5511d6b0fc6c-1200x675.webp?w=600&q=75&auto=format)
Technology AI Voice Detector: How to Detect AI-Generated Audio
62% of organizations told Gartner in 2025 that they had faced a deepfake attack in the previous 12 months. Here's how AI voice detectors work, 5 tools compared, how to test a detector on your own audio, and how to spot synthetic speech by ear.
Cybersecurity AI and LLM Security: Every Major Threat Vector Mapped
AI and LLM security spans prompt injection, supply chain attacks, agent control flow vulnerabilities, and model misuse. This guide maps every major threat vector and links to in-depth breakdowns so you can defend your AI systems today.
Cybersecurity AI Agent Attack Surface Map and Security Checklist
The first developer-friendly attack surface map combining OWASP's Top 10 for Agentic Applications, Cisco's MemoryTrap disclosure, and June 2026 red-teaming benchmarks showing 70% attack success rates — with a printable security checklist.
Cybersecurity Advanced Prompt Injection Techniques 2026: 7 Attack Chains Beyond OWASP #1
Prompt injection graduated from academic curiosity to active exploit — with CVEs filed against GitHub Copilot, Claude Code, Cursor, and AWS Kiro in a single month. Here are the 7 advanced attack chains researchers are tracking and the only defense architecture with provable security.
Showing the 30 newest of 34. See the full AI Agent Architecture & Frameworks pillar for the rest.
Keep exploring
Close to this topic, not on this page.
