security on Saurabh Kumar

Rampart

Sun, 11 Oct 2026 12:00:00 +0000

Small on-device models are getting good enough to sit in front of the big ones. Rampart from the National Design Studio redacts PII in the browser before a prompt reaches any LLM: regexes with checksum or structural checks for cards (Luhn), SSNs, emails, URLs and IPs, plus a fine-tuned MiniLM for names, phones, account numbers, government IDs and street addresses.

14.7 MB Q4 ONNX (~15 MB with tokenizer), 98.4% private-term recall across seven Latin-script languages. p50 is 3.9 ms on WebGPU but 12.6 ms on the default WASM backend. OpenAI’s Privacy Filter is ~2.8 GB; Presidio scores 65% on the same test set.
guard.protect() swaps PII for placeholders like [GIVEN_NAME_1], and guard.reveal() fills them back into the reply. The LLM never sees the name; the user still does.
Default-deny: every detected label is redacted except city, state and ZIP, so the assistant keeps coarse location without the street line.
Know the gaps: names in non-Latin scripts recall ~14% (Hindi 6%, Arabic 5%), and it scores 86.4% on a hostile-input suite. The repo calls it harm reduction, not perfect protection.
Same size class, different job: Cactus Whistle is 16.9 MB speech-to-text in seven languages, 11 ms to first token on an M4 Pro CPU (Whisper base: 145 MB, 73 ms). It loads into the same engine as Needle 3, their tool-calling model, so one call takes audio in and returns set_lights(room="kitchen", on=false). Voice commands for home automation with no cloud round trip.
The pattern: a tiny specialist at the edge does one narrow job (redact, transcribe, route), and the big model only sees what it needs. Both sets of benchmarks are vendor-reported.

Microsoft eXecution Container (MXC)

Sat, 10 Oct 2026 12:00:00 +0000

Microsoft eXecution Container (MXC) is an embeddable SDK for running untrusted commands under OS-enforced policies, not a hosted sandbox service.

Defaults: Seatbelt on macOS, Bubblewrap on Linux, ProcessContainer on Windows. SDKs for Node, Rust, and .NET.
Restricts filesystem and network access. macOS hostname allowlists need a caller-managed proxy; not every backend is a VM.
Useful boundary: sandbox the command where it executes, not just the agent’s tool-composition script. Isolation limits damage; it does not make generated output trustworthy.

SecretSpec

Wed, 15 Jul 2026 12:00:00 +0000

SecretSpec declares secrets once in secretspec.toml (names + descriptions, never values) and resolves them from any of 11 backends: keyring, 1Password, Vault, AWS, GCP, .env. Same command everywhere: secretspec run --profile production -- npm start. Per-secret fallback chains (providers = ["vault", "keyring", "env"]) and a ref = { item, field } form points at a secret you already have elsewhere, no renaming. Sharpest bit: agents get held to a stricter bar by default. require_reason = "agents" forces secretspec run --reason "..." before an AI agent can read a secret, appended to a local audit log; humans running interactively are unaffected. Rust core, SDKs for Python/Go/Node/Ruby/PHP/Haskell. Worth piloting on a project drowning in .env sprawl.

Opus 4.6 launch

Fri, 06 Feb 2026 12:00:00 +0000

Opus 4.6 launch.

context compaction (beta) and 1M context window, enables longer agentic tasks without loosing context.
they claim it has found 500 Zero-Day Flaws in open-source projects (yet to see the proofs though)
agent teams, multiple agent coordinates with a leader agent. https://code.claude.com/docs/en/agent-teams

UIPC and SIPC on blockchain

Thu, 05 Feb 2026 12:00:02 +0000

On blockchain, UIPC (Universal IPC) and SIPC (Secure IPC) are frameworks that allows different network to talk to each other using a standard protocol. While UIPC acts as the universal courier, SIPC acts as the security guard. Together, they create “Internet of Blockchains”. Gemini

Cosmos IBC & Polkadot - XCM are two industry leading protocols for UIPC
For the SIPC, Zero-Knowledge Proofs and Rollup architecture also “Trustless bridges” or “Cross-rollup communication” are often more frequently used terms in blockchain literatures.

Handling Django Authentication Redirects in HTMX Applications

Fri, 02 May 2025 00:00:00 +0000

When a user’s session expires, an HTMX request can follow Django’s login redirect and swap the login page into the target element. I ran into this recently and used middleware to redirect the whole browser instead.

What’s Going Wrong?

HTMX updates part of a page, but an expired session can cause Django to redirect a request to the login page. The returned login HTML can then appear inside a div or table cell rather than replacing the page.

What we really want is to redirect the whole browser to the login page, and then bring users right back to where they were after they log in again. Simple idea, slightly tricky execution.

Redirect with middleware

This middleware converts HTMX 302 responses into a full-page redirect using HX-Redirect. It sets a next parameter from the referring page or request path.

from urllib.parse import urlparse


class HtmxAuthRedirectMiddleware:
 """
 Middleware to handle HTMX authentication redirects properly.

 When an HTMX request results in a 302 redirect (typically for authentication),
 this middleware:
 1. Changes the response status code to 204 (No Content)
 2. Adds an HX-Redirect header with the redirect URL
 3. Preserves the original request path in the 'next' query parameter

 This ensures that after authentication, the user is returned to the page
 they were attempting to access, maintaining a seamless UX with HTMX.

 Credits: https://www.caktusgroup.com/blog/2022/11/11/how-handle-django-login-redirects-htmx/
 """

 def __init__(self, get_response):
 self.get_response = get_response

 def __call__(self, request):
 response = self.get_response(request)
 # HTMX request returning 302 likely is login required.
 # Take the redirect location and send it as the HX-Redirect header value,
 # with 'next' query param set to where the request originated. Also change
 # response status code to 204 (no content) so that htmx will obey the
 # HX-Redirect header value.
 if request.headers.get("HX-Request") == "true" and response.status_code == 302:
 # Determine the next path from referer or current request path
 ref_header = request.headers.get("Referer", "")
 if ref_header:
 referer = urlparse(ref_header)
 next_path = referer.path
 else:
 next_path = request.path

 # Parse the redirect URL
 redirect_url = urlparse(response["location"])

 # Set response status code to 204 for HTMX to process the redirect
 response.status_code = 204

 # Update the "?next" query parameter
 query_params = parse_qs(redirect_url.query)
 query_params["next"] = [next_path]
 new_query = urlencode(query_params, doseq=True)

 # Set the new HX-Redirect header
 response.headers["HX-Redirect"] = f"{redirect_url.path}?{new_query}"

 return response

This middleware is inspired by this post by the caktus group.

How It Works

The middleware changes a 302 response to 204 so HTMX can process the HX-Redirect header. That header redirects the whole browser. The next parameter records the path to return to after login.

This applies to all HTMX 302 responses, not just authentication redirects. Add checks if other redirects in your application need different handling.

Adding This to Your Project

Add the middleware to your Django project and its MIDDLEWARE list in settings.py:

MIDDLEWARE = [
 # Your other middleware stuff...
 'yourapp.middleware.HtmxAuthRedirectMiddleware',
]

Just make sure it comes after Django’s authentication middleware in the list - order matters here!

Edits:

2025-05-04: Use urlparse and parse_qs to securely handle urls. Thanks Adam Johnson

actions/checkout

Tue, 10 Dec 2024 12:00:00 +0000

actions/checkout saves the workflow token (GITHUB_TOKEN or a custom one) in the checked-out repo’s .git/config by default. Workflows that upload the repo as an artifact leak it, and Unit 42 found many projects doing this. GitHub calls it intended behaviour. Turn it off, and scan workflows with zizmor, which would have caught the Ultralytics injection.

- uses: actions/checkout@v4
 with:
 persist-credentials: false