
I wish the people releasing early pocs to one of the most severe RCEs in history a very fuck you
A moron has released a POC early for php2shell so here is what you need to do:
If you’re on version prior to 8.2:
Edit php.ini: session.upload_progress.enabled = Off
Then reload php-fpm and restart apache
If you’re on later version:
Update (it’ll be available in a few hours)
In any case, the mitigation for the php.ini should work either way. Not sure if it’ll cover both vulns though so keep an eye on that. I’ll be reporting on follow up activity on windows as a target.
Good luck!
If we hadn’t already created cloudflare.config.ts this would have been a cool way to solve generating env types from a wrangler file 😅
Agents have gotten so good that they've become so tolerant of tech debt (ie. slop) while being trained to non-vocal about it. As models become better these days of reckoning when the model can no longer push past the tech debt will become rarer but much more severe.
The nice part about building on Durable Objects is that you get to surprise your customers like this
One day there will be people that only know Cloudflare as an AI lab and then get confused when we tell them we also have a cloud
a brand new model paradigm! introducing 𝚌𝚕𝚊𝚒𝚛𝚒𝚌: a system one model for extracting, computing, estimating, and predicting values from unstructured inputs.
blog.cloudflare.com/clairic-…
It’s really sad shit like this is getting increasingly normalized in our industry. Even a year ago when this platform was full of this kind of commentary you could take solace in the fact it was rare on tech twitter.
I can’t go 15 minutes without seeing a slur in comments
Don’t blindly follow benchmarks and don’t blindly follow what people say about any given model on here, go actually use them yourself and see how they do in your own workflow.
I’ve seen people glaze gpt 6 for it to work poorly for me and people shit on Gemini 4 for it to be fine
It's a shame Anthropic and OpenAI sub don't really give you programatic access to other models like text-to-speech and speech-to-text, so i've just been telling my agents to use the `cf ai run` command with *fantastic* results
Everyone should be doing this with any non-trivial plan. LLMs are terrible at choosing structure and abstractions, but great at picking out the important bits of existing code.
nobody reads pull request descriptions, so we made videos
when @polylane submits a PR to you, we attach a video walkthrough explaining exactly what the issue is, and how it fixed it
@viviane_trn wrote down how we built it here: polylane.com/blog/nobody-rea…
Gemini 4 is such a happy lil guy, the writing style is like the opposite of the gpt models.
I’m blowing through my context though, not sure if this is just a symptom of using opencode though
GLM-Legal uncensored:
Introducing GLM-Legal
Built on GLM-5.3-Flash, GLM-Legal matches frontier models on mainstream legal benchmarks.
Designed for real legal workflows: contract review, evidence analysis, compliance, and more.
Tech Blog: z.ai/blog/glm-legal
Apply for API Access: z.ai/consultation/legal
Try AutoClaw: autoclaw.z.ai
zeb retweeted
two billion events per second
Amazing blogpost from the Data Team here at @Cloudflare on how we optimized our ClickHouse platform and gained 2.5 YEARS of hardware runway.
If you've tried to by hardware this year, you know that's a huge win!
And as always, we're hiring :)
blog.cloudflare.com/rewindin…
zeb retweeted
if you squint just a little bit Cloudflare DOs are not unlike Erlang/Elixir processes...
...and this sort of makes the Cloudflare platform like a planet-wide BEAM cluster.




















