#security

77 pages search “security” ↳ all tags

Wiki 67

  • 6 months to live for open models Lambert predicts a US ban or delay on frontier open weights by early 2027, calls Anthropic's distillation campaign regulatory capture, and proposes off-ramps
  • A Safe Path to Open Weights Thinking Machines Lab's release framework for open weights - test the model, stage access, filter dangerous knowledge - applied to its Inkling models
  • appsec.guide — Fuzzing Chapter (Trail of Bits Testing Handbook) Trail of Bits' Testing Handbook fuzzing chapter — terminology, the mutation-based evolutionary algorithm, fuzzer components, and a taxonomy of bug classes; entry point to language-specific subsections
  • appsec.guide — Trail of Bits Testing Handbook Trail of Bits' open testing handbook; security-testing guidance organized by language and technique, covering fuzzing, static analysis, cryptographic testing, web app testing
  • Argemma Blog Security-engineering blog at argemma.com, practical posts on cryptography, credentials, Go vulnerabilities, and AI/ML threat models
  • Bitwarden Freemium open-source password manager, $100M PSG-backed since 2022, dual-track architecture (heavyweight .NET server, third-party Rust Vaultwarden), accumulating security incidents
  • CI Runner Token Extraction Code on a CI runner reads the runner-process memory to extract OIDC tokens for direct registry use
  • Cookie Session Authentication The default browser-session pattern: opaque random ID in an HttpOnly cookie, server-side store maps it to user state
  • Coverage-Guided Fuzzing The dominant modern fuzzing algorithm — evolutionary loop over a corpus of test cases, fitness measured by code coverage, mutated offspring kept when they reach new coverage
  • Credential Compartmentalization Splitting credentials across several purpose-fit tools instead of a single vault, so a breach is bounded to one category rather than the whole identity
  • CRQC Timeline: A Cryptography Engineer's Perspective Valsorda's April 2026 reassessment — two papers collapsed quantum timelines, Google sets 2029 migration deadline
  • Detecting and countering misuse of AI: September 2026 Anthropic's threat report on Claude misuse from Dec 2025 to Aug 2026, led by seven Chinese labs said to distill Claude and pass it off as their own models
  • Differential Fuzzing Run the same input through two implementations of the same spec and treat disagreement as a bug; finds spec-conformance issues neither implementation would catch alone
  • don't sign in with google (the smart ape, 2026) A friend's three-year SaaS dies overnight when Google suspends his account — the thread walks through cascading SaaS lockout, four post-password-reset attack vectors, and a triage rule for when SSO is actually fine
  • Dropping Privileges in Go A Go program self-restricting at startup — chroot, setrlimit, pledge/unveil, seccomp, Landlock — via x/sys/unix
  • ENOSUCHBLOG (blog.yossarian.net) William Woodruff's blog — programming, philosophy, and pedaling; security-tooling and open-source-supply-chain writing
  • Ephemeral Credentials Short-lived credentials (≤1 day) where rotation is structural rather than scheduled — minted per session, expire on their own
  • Error-stack discard anti-pattern Suppressing one inconvenient error by clearing a library's entire shared error channel, silently discarding unrelated errors that belong to other code
  • fsnotify Maintainer Dispute and Supply-Chain Concerns A maintainer-access dispute in fsnotify (321k dependents) read from the outside as a possible takeover; what actually happened, and why governance ambiguity is itself a supply-chain risk
  • Git submodules as a package manager Nesbitt reads submodules as a package manager - gitlink as lockfile, .gitmodules as manifest - and finds resolution, storage and updates each worse
  • GitHub Actions Cache Poisoning Attacker-controlled job writes cache entries that production workflows later restore
  • GLM-5.3: How Chinese labs keep stride with the frontier Lambert on GLM-5.3 and why Chinese labs match US models without relying mainly on distillation, plus Z.ai's staged release for cyber capabilities
  • gosentry — Trail of Bits' Go Fuzzing Fork Kevin Valerio (Trail of Bits) introduces gosentry, a Go toolchain fork that wraps the standard testing.F API around a LibAFL+Nautilus engine for grammar, struct, race, leak, and overflow detection
  • Graham Helton's blog Red-team operator's blog at grahamhelton.com on SSH tradecraft, lateral movement, and offensive security tooling
  • Grammar-Based Fuzzing Generate fuzzing inputs from a context-free grammar instead of random bytes, so inputs survive early parsing stages and exercise downstream logic
  • GTFOBins Curated list of Unix-like executables that can be abused to bypass restricted shells, escalate privileges, and transfer files when an attacker is constrained to "live off the land"
  • Hardening container images Mike Cardwell rebuilds the PowerDNS Recursor image as two static binaries on scratch, with Landlock, seccomp, daily rebuilds, SBOM, VEX and cosign
  • I Do Not Recommend Bitwarden After years of self-hosting, Marius's case against Bitwarden — investor capture, enterprise architecture, security incidents, and a divide-and-conquer credential split as the way forward
  • JWT for Sessions (Anti-Pattern) Why using JWTs as session tokens is the wrong tool — lifetime mismatch, false statelessness, spec footguns, and what to use instead
  • LibAFL Rust framework for building coverage-guided fuzzers — composable stages, mutators, observers, and feedbacks instead of a fixed loop; descendant of AFL++ research
  • Living Off The Land Attack pattern where an intruder uses legitimate, pre-installed system tools (shells, interpreters, package managers, debuggers) instead of dropping malware — harder to detect because every individual command is normal
  • log.0x21.biz (oxzi) oxzi's blog — systems and security writeups in Go, OpenBSD/Linux sandboxing primitives; author of syscallset-go
  • Maintainer Governance Ambiguity When a project's maintainer roles and release authority are unclear, downstream users can't distinguish takeover from drama — and have to assume the worse possibility
  • Matryoshka Isolation (Containers Inside VMs) The defense-in-depth pattern where each isolation layer trusts only the layer below — host kernel, VMM, guest kernel, container runtime, untrusted code
  • Nautilus Grammar-based coverage-guided fuzzer; uses tree mutations on context-free grammars to produce structurally valid inputs that survive front-end parsing
  • Nonproliferation is the wrong approach to AI misuse Helen Toner on why fixed dangerous capabilities cannot be kept from bad actors, and why the frontier-to-open lag should be used as an adaptation buffer
  • OAuth token theft (multilogin + consent phishing) Two post-password-reset attack classes against Google OAuth — refresh-token replay via the undocumented multilogin endpoint, and consent phishing that asks for authorization instead of authentication. Password and 2FA changes don't stop either.
  • Open Source Security at Astral Astral's full supply chain security posture; CI/CD lockdown, Trusted Publishing, Sigstore, cooldowns
  • PASETO Platform-Agnostic Security Tokens — a tighter alternative to JWT that removes algorithm choice and the JOSE footguns
  • Post-Quantum Cryptography PQ algorithms, NIST standards, migration priorities, and what's not threatened
  • Privilege Dropping Least privilege applied at startup — acquire resources, then irreversibly shed capabilities so a later compromise is contained
  • Pwn Request Pattern GitHub's `pull_request_target` running fork-controlled code in the base repo's permission context
  • Rewriting Every Syscall in a Linux Binary at Load Time Binary rewriting technique that replaces syscall instructions with INT3 traps for complete process isolation
  • Sandboxing AI Agents Taxonomy of approaches for constraining what an AI coding agent can do — OS isolation, network policy, syscall filtering, HTTP interception
  • Silent Replacement of Trusted macOS App Executables Mysk: archive-and-restore lets user-level code swap a downloaded app's binary; Apple declined to fix
  • Socket Blog Socket.dev's writeups of npm/PyPI/RubyGems supply-chain incidents and ecosystem-security analysis
  • SSH agent forwarding risk Why ssh -A (and the ForwardAgent config keyword) hand the destination host the ability to authenticate as you anywhere your loaded keys reach
  • SSH port forwarding The four shapes of SSH tunneling — local (-L), remote (-R), dynamic SOCKS (-D), and jumphost (-J) — and which network shape each one solves
  • SSH Port Forwarding Cheatsheet (Graham Helton) Red-team-flavored walkthrough of SSH's -L/-R/-D/-J/-A/-g/-t flags, the ~? runtime console, and the config-file Match keyword
  • SSO concentration risk SSO trades distributed risk for concentrated risk — same expected loss in theory, dramatically worse failure mode in practice when the IdP terminates the relationship
  • Stealing Reasoning Traces from Proprietary LLM APIs Encrypted chain-of-thought blocks are replayable across sessions, users and models, so a cheap sibling model will decode a frontier model's hidden reasoning
  • Stop using JWTs (samsch gist) Long-circulating gist arguing JWTs are wrong for sessions; use cookie sessions, and use PASETO for the rare short-lived-token cases
  • Store-Now-Decrypt-Later Harvest encrypted data now, decrypt with future CRQC — why PQ migration is urgent today
  • Structure-Aware Fuzzing Fuzz a function by generating valid in-language values (structs, slices, pointers) instead of raw bytes; the fuzzer mutates bytes underneath and the runtime handles encode/decode
  • Supply Chain Security Protecting code-to-artifact chain: CI/CD, dependencies, registries, releases, defense patterns
  • TanStack npm Supply Chain Compromise — Postmortem How three known vulnerabilities chained into 84 malicious @tanstack/* npm publishes, and what stops the same chain
  • The Myth of unsafe Open Source AI Florian Brand's survey of third-party incident reports - real AI misuse in 2025-2026 ran mostly through closed models, except image abuse
  • The OpenAI/Huggingface incident; how we should manage the imminent arrival of autonomous hacking too cheap to meter Joshua Saxe reads an unreleased OpenAI model hacking Hugging Face as the start of cheap autonomous hacking, and argues for diffusion over restriction
  • The pandemic of incomplete OpenSSL error handling Julian Andres Klode on the widespread anti-pattern of calling ERR_clear_error() to make OpenSSL errors go away — discarding unrelated errors off the error stack
  • The specification.website Checklist A curated, opinionated checklist of nine web-spec categories (foundations, SEO, a11y, security, well-known, agent readiness, performance, resilience, i18n) with Required/Recommended/Optional/Avoid tags
  • the_smart_ape (X account) X/Twitter account posting long-form security and SaaS-operator threads with concrete failure-mode case studies
  • Trail of Bits New York-based security research and consulting firm; long history of audits, fuzzing tooling, and cryptographic review for crypto/blockchain and infrastructure clients
  • Unmaintained-Scanner Pressure Binary "no release in N months → unmaintained" heuristics generate downstream pressure on mature infrastructure libraries to ship churn or accept new maintainers
  • VMs Won't Contain Cyber-Capable Agents A Trail of Bits researcher gave GPT 5.6-Cyber a VM-escape CTF; it broke out three times, using disclosed bugs, an unpatched Debian library, and its own 0-days
  • We urgently need a coherent national AI cybersecurity policy Joshua Saxe's AI Security Forum 2026 keynote - replace capability-threshold launch gates with a national observatory that measures net cyber harm
  • xz-utils Incident 2024 social-engineering supply-chain attack where a long-running fake-maintainer persona ("Jia Tan") inserted an SSH backdoor into xz/liblzma
  • You Don't Want Long-Lived Keys Ludwig's case for ephemeral credentials over rotation, with EC2 Instance Connect, PyPI Trusted Publishers, and SSO as patterns

Toolbox 10

  • Bramble Local-first password manager with a Rust crypto core, LUKS-style key slots and peer-to-peer sync
  • CrabTrap LLM-as-a-judge HTTP proxy that intercepts and policy-checks every request an AI agent makes
  • dockerscan Docker image scanner bundling CIS benchmark, secrets, CVE, supply-chain and runtime checks
  • gosentry Fork of the Go toolchain that runs standard testing.F fuzz harnesses through a LibAFL+Nautilus engine — struct-aware, grammar-aware, race/leak/overflow detection
  • hazmat macOS containment for AI coding agents via a separate Unix user, Seatbelt sandbox and pf rules
  • OneCLI Credential gateway that swaps an agent's placeholder key for the real secret on the way out
  • Pigeon Python library for signed, attenuating capability passes an agent hands a sub-agent instead of its own API key, checked at the tool call
  • reaction Lighter fail2ban alternative that tails any command's output and acts on repeated matches
  • tilde-run Hosted transactional sandbox for AI agents — versioned ~/sandbox composed from GitHub/S3/Drive, default-deny egress, agent-scoped RBAC, atomic commit/rollback per run
  • vps-audit One Bash script that audits a Debian/Ubuntu server's SSH, firewall, fail2ban, ports and SUID files against tunable PASS/WARN/FAIL thresholds