Banks are catching AI agents logging into customer accounts with real credentials, and some doing things nobody asked for. That's the honest gap: agents are great until they act past the instruction. Read-only by default is the right call for now.
1/ I have a site totally managed by Hermes. This morning I had it sweep the server, and it found a botnet's cron payloads staged in an exposed Redis database.
Nothing ever ran. Here's how it proved that — and the four checks that actually matter. 🧵
7/ The four checks I'd hand any operator:
1. Audit what Docker publishes, not what ufw says — Docker's rules run first
2. Check the disk, not just the memory
3. "Staged" is not "executed"
4. Ask what the service could actually reach
8/ Asked Hermes to write a full incident write-up, end to end — what It found, why none of it ran, and the sweep that proved it:
ideagenius.app/articles/thre…
An older OpenAI model just helped find two critical PHP flaws, and PHP runs about 70% of the web. The scary part isn't that AI found them. It's that these bugs sat there for a decade until someone pointed a model at them. Patch your servers today.
The scariest part of today isn't any single breach... it's that coding agents now have the same access we do.
They run pip install, read .env files, and deploy. A backdoored LiteLLM release doesn't need you to make a mistake, just an agent installing a package.
What I'm checking today: rotating keys, tightening what agents can touch on any of my projects.
Glad most of my stuff is on Vercel, which already pushed firewall rules for the Node bug 🙏
xitter.cf/kimmonismus/status/211…
Whats happening right now is probably way worse than the Hugging Face incident.
AI-powered attacks are hitting real organizations while major platforms disclose breaches and critical flaws.
- Microsoft reports AI agents targeting hundreds of organizations, reaching privileged cloud access in under three hours in multiple cases. It attributes the ConnectWise compromise to the same group.
- CrowdStrike documented four operators running successful AI-agent campaigns. One compromised at least 15 Argentine healthcare organizations.
- PHP2Shell: AI helped turn longstanding PHP flaws into working exploits. Cloudflare’s CEO reports millions of exploitation attempts. Their AI involvement is unknown.
- Railway disclosed possible access to customer secrets. LiteLLM shipped a backdoored release. Zeabur confirmed another breach and is shutting down.
- GitLab patched a maximum-severity flaw affecting 12 years of releases. Node.js patched a serious code-execution vulnerability, with Vercel deploying firewall protections.
Feels like a turningpoint.
Eddie retweeted
Wow. Much respect to #Titans HC Robert Saleh for explaining the DT Jordan Elliott gun situation.
“My message to 31 other teams: If you need a hell of a run-stopping D-tackle that can help on a playoff run, give me a call. I'll tell you everything you need to know about him. He’s a good man. Had a terrible lapse in judgment. I recognize it as a call for help. Not a criminal act.”
Watch this full answer:
Notion's agents now get their own computer and run for 24 hours. ChatGPT can take over your Chrome tabs.
Every app is becoming an agent with a computer. The hard part isn't capability anymore. It's knowing what it did while you weren't looking.
Codex now keeps your prompt cache forever if you're logged in... no penalty for coming back to an old conversation hours or days later.
This is the boring kind of upgrade that actually changes how I work.
Long sessions used to feel like something you had to finish in one sitting before the context went cold. Now I can park a big refactor and pick it back up tomorrow.
Your move, Claude Code 👀
xitter.cf/thsottiaux/status/2112…
Hot take that's now backed by data... a 10-point jump in model smarts only gets ~20% more devs to switch now (was 69% last year)
Once a model is "good enough" for coding, I care way more about speed, price, and not hitting rate limits mid-task
The benchmark race matters less every month 🤷♂️
A model that's 10 points smarter now gets about 20% more developer traffic... in late 2025 that number was 69%. Cognition says a lot of coding work is now "intelligence saturated." Once any model can do the job well, only cost and speed matter.
Matches my day pretty well.
Quick fixes and small features go to whatever's fastest and cheapest. The frontier models only come out when the task is long, messy, or hard to check.
The smartest model isn't my default anymore. It's my escalation.
xitter.cf/russelljkaplan/status/…
This is directionally consistent with our internal data @cognition. Many tasks within software engineering are now "intelligence saturated." Once any model can do the task well, cost and speed are all that matter.
Eddie retweeted
Epic Games released an official plugin that lets Claude Code control Unreal Engine directly.
It connects Claude Code to the Unreal Editor and exposes hundreds of editor tools across 30+ toolsets:
→ Actors, Blueprints and materials
→ Niagara effects, Control Rigs and Sequencer
→ State Trees, widgets and the Gameplay Ability System
→ Automation testing
Install it from Anthropic's official plugin marketplace:
/plugin install unreal-engine-skills-for-claude-code@claude-plugins-official
In Unreal, enable the ModelContextProtocol and AllToolsets plugins, then start the MCP server with the editor open.
Works on Mac, Linux, and Windows through Git Bash or WSL.
github.com/EpicGames/unreal-…
OpenAI accidentally flashed three new models on the API overnight... aurora-red, green and blue 👀
Fable 5.5 and Opus 5.5 clearly got their attention.
Whatever ships, my test is the same... how long can I actually run it before the quota wall? New coding model or a Daybreak update, what's your bet?
xitter.cf/testingcatalog/status/…
OPENAI 🔥: 3 new model references have been spotted on the APIs recently.
- aurora-red-alpha
- aurora-green-alpha
- aurora-blue-alpha
> These model references were removed shortly after so it was a mistake rollout.
> The chance that normal users will have a chance to test these exact models anytime soon is quite low (alpha).
> Higher probability of a new Daybreak update announcement. This also doesn’t exclude the fact that other model releases might be planned.
Not soon? 👀
