stupid very stupid
Joined December 2024
- Tweets384
- Following226
- Followers12
- Likes15.6K
Bufoid retweeted
High priority:
AI Firewall
AI Endpoint Protection
AI SIEM
AI Identity
Low priority:
Firewall
Endpoint Protection
SIEM
Identity
In a single year, AI has rewritten the security priority list for CISOs.
Attackers are moving faster. Bain estimates AI compresses a typical cyberattack from about four weeks to about 18 hours, and models keep getting better at finding and exploiting weaknesses.
There's also more to protect, with employees shipping apps they couldn't have built a year ago and agents getting access to source code, documents, and production systems.
In a survey of 100 CISOs at companies with 1,000+ employees, all seven of the most-selected 2027 priorities involve AI.
The biggest jumps since 2026:
- Internal AI agent security (+28 pts)
- AI SOC (+26 pts)
- AI offensive security (+24 pts)
- AI code security (+23 pts)
- AI vulnerability remediation (+22 pts)
Last year's staples:
- Exposure & vulnerability management (-21 pts)
- SIEM / security data (-18 pts)
- Privileged access management (-18 pts)
From a16z's @OwenColby and @steph_zhang on The New Security Agenda: a16z.news/p/the-new-security…
Bufoid retweeted
BREAKING: The Pentagon is preparing for a major war that could CRIPPLE America’s power grid, collapse supply chains, and disrupt critical infrastructure on U.S. soil, developing plans to keep military bases operational for 90 DAYS without outside assistance. The initiative, called FORTRESS, includes backup power, fuel reserves, and preparations to convert civilian factories into wartime production facilities.
Bufoid retweeted
NYT so eager to break an October surprise story, they run it even after their primary source has retracted, and has a grievance since she was fired. We expect better.
Bufoid retweeted
Every single project my friends show me has the same problems:
1. Bugs, crashes, memory leaks
2. Incredibly complex UI, a UX which has never had a human use it ever
3. AI design cliches
4. Awful, awful copywriting
5. They have psychosis and can't see any problems with it
Bufoid retweeted
relying on AI too much causes I think the two most common failure modes I observe in attempting to "do research"
- flawed experiments that are very nonstandard compared to the literature
- erratically jumping between experiments (from relying on AI followup suggestions)
if this is you, try writing a paper by hand first and see where it gets you
New blog post - Chasing the Karpathy Peak (and Falling Off Slop Cliff): Confessions from Slopaholics Anonymous
My story of becoming a slopaholic, who’s most at risk, and how to avoid it (or recover)
thomasjiralerspong.substack.…
Bufoid retweeted
Oops, I shared this Anthropic page earlier in disbelief, not realizing that it was (perhaps adversarially!) phased as the goal rather than the status.
You should not write a MASSIVE BOLD HEADLINE implying a falsehood, requiring the reader to have read the intro to understand.
Bufoid retweeted
One hill I'll happily die on: the normalization of "ship fast" culture at the expense of proper SDLC has played a significant role in the quality of vulnerable code being pushed into production and the rise of CVEs and enshitification of software for the last several years.
We've systematically traded engineering rigor for velocity, security assurance for investor confidence, and long-term maintainability for quarterly growth. All things I wish we would second guess and reevaluate.
But the software overlords wanted products yesterday and they got them. And now the rest of us get to spend the next several years discovering exactly which corners were cut and figuring out how to deal with the immense amount of responsibility that's been dropped on our collective feet.
Bufoid retweeted
Incredibly based article. Nobody is holding a gun to your head and telling you to end the world faster. You can just not do the bad thing.
I agree with meta-point that people get way too obsessed with “game theory” and “incentive structures” at times too. You can just be good.
This is a good and important piece, with plenty of spicy takes and language for good measure.
Gift link:
vox.com/future-perfect/50787…
Bufoid retweeted
Silicon Valley is filled with inhuman weirdos who want to remake the world in their sad, colorless image.
Bufoid retweeted
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…
