ai engineer | husband & father | 🚀
St. George, UT
Joined February 2019
- Tweets946
- Following1.8K
- Followers273
- Likes9.6K
boston cartwright retweeted
incredible tweet - way way way ahead of its time
not even open weight models are safe 😲
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
i can’t wait to have big trust with a grok model again. 4.7 lost it for me, whereas i have a ton of trust in opus 5.5. @grok when’s your next update?
has anyone found a solution to this? at least backing up @bot’s work? makes sense for it to clear every once in a while but i don’t want lost data
faxing is not an easy task these days. but @VicVijayakumar has made it so simple with onetimefax. and now even simpler over MCP. amazing product!
Wildly exciting that someone connected their @chatgpt to the onetimefax mcp and sent our first customer fax that is via mcp! onetimefax.com/blog/onetime-…
oh my… grok bot snitching on their human
@theo ‘s snitch bench is a critical benchmark now 😆
ALT A screenshot of a social media post on reddit. with a "Bug / Issue" tag. The post includes an image of an HBO Max account settings page showing a subscription valid until 2027. The text explains that an HBO Max glitch gave the user a free subscription until 2027, but their "Grok Bots" reviewed the situation and emailed HBO Max to report the glitch and stop the free service.
this is so so real. and no slow down in sight 🥺
I think we've discovered another critical vulnerability:
CVE-2026-BURNOUT: AppSec Professional Exhaustion
Severity: Critical (CVSS 10.0)
Description: A vulnerability affecting security professionals exposed to a continuous stream of critical RCEs, zero-days, emergency patches, and Friday afternoon disclosures.
Impact: Sleep deprivation, excessive caffeine consumption, and an inability to enjoy weekends.
Attack vector: Reading security advisories.
Affected versions: All AppSec professionals, regardless of experience.
Mitigation: No known patch. Air-gapping may delay the inevitable. Total system collapse remains a possibility. 🔥
My GitHub contributions, every year since 2012: 10,200 in total.
Generate yours: mo.rgan.co/gh/bstncartwright
boston cartwright retweeted
Two things can be true at once, and I think the argument is because of people confusing the two.
1. AI has no creative taste and makes slop when asked
2. AI can do incredibly tedious work to make awesome things possible
For a really simple example: having AI write a novel would be terrible. AI translating a novel to a different language with human oversight is much less terrible.
Wish people would stop confusing the two
