Manan πŸ€¦πŸ½β€β™‚οΈ (@manan) header
Manan πŸ€¦πŸ½β€β™‚οΈ (@manan) profile picture

Cybersecurity stuff. Wrote tech for ZDNet, TNW, others. Created @cyberprismapp, @everythingms, @fussyvegetarian @artfallapp

Seattle & Bombay
Joined January 2008
Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)Photo by Manan πŸ€¦πŸ½β€β™‚οΈ (@manan)
hey @sama heard you want everyone’s PC RAM? I’ll give you all of mine, 1 OAI share for each 8GB…
3
226
the only reason I find this problematic is because I am not making $200M at their IPO…
We’ve trained Claude to flag potential misuse in conversations, a technique we call self-assessment. Right after it responds, Claude gives a preliminary safety assessment of the conversation. Since we began using self-assessment in June, it has surfaced misuse that our other safeguards weren’t designed to catch. Because Claude has already read the conversation, self-assessment is very efficient. And unlike classifiers, which target a discrete set of harms, it draws on Claude’s own judgment, so it gets more accurate as our models improve. Read more on our new Safeguards blog: safeguards.anthropic.com/202…
1
88
Manan πŸ€¦πŸ½β€β™‚οΈ retweeted
Black October just got its final boss: with the largest crypto heist in history. In one month : 1. Pentagon personnel data breach (3M people) 2. Denmark population register breach (8.8M people) 3. French supermarket freezers remotely disabled 4. Ahold Delhaize grocery systems knocked offline 5. Educational Testing Service California student data theft (5.7–7M) 6. eMudhra certificate authority key compromise 7. ConnectWise Cloud Backup supply-chain attack 8. PHP2Shell critical RCE flaws (CVE-2026-109618/109619) 9. LiteLLM malicious package supply-chain attack 10. Railway environment variables breach 11. Zeabur hosting platform breach (3.2M records) 12. GMO Onamae Desktop Cloud malware (26k+ servers) 13. Oracle Health/Cerner patient data breach (20M) 14. Capcom ID account exposure (11.7M) 15. University of New Mexico Health ransomware disruption 16. Quest Diagnostics lab platform breach 17. University of Copenhagen data breach (525k) 18. DHS Icepick employee records leak (41k+) The big boos is 19. Binance hot wallet drain ($5.5- 6.3B) largest crypto heist ever
15
88
14
495
21,594
please, again, who are these CISOs that a16z is talking to for this stuff?
In 2027 CISO's will be paying for the steak dinners.... Never been a better time to CISO!
72
Who are these CISOs??? - last week, vuln and attack surface management is going to be crazy 12-18 months - this week, well yea no I’m worried about AI spending… yea
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…
1
1
56
keeping battery saver on my iPhone is very effective. Why is the regular mode not the battery saver mode
95
kahani ghar ghar ki
Watch Ambanis also face the same struggle as yours... Hope it helps
76
walking around Germany & the Meta glasses are very useful QoL upgrade. real time translation has been a bit of miss: - People breaking into partial English - ambient discussions being picked up - when I ask it to tell me what I should say in German, it speaks so quick that it’s useless
1
72
one other thing that’s been annoying is that when I am talking to the Meta AI -> AI is working on the answer -> my kid says something or I turn around to say something to wife / kids -> Meta AI takes that partial thing as new context
1
17
very very effective and helpful for: - self checkout screens at stores - menus - photos, hands-free - taking calls - asking random questions about where I am (Did have to reluctantly give Zuck access to all the time precise location while I needed the maps stuff, disabled it after)
18
hi @UberEats, great use case for AI. I walk english, I talk English, I laugh English… your app can translate ALL of this with an LLM into English for me when I am travelling to a country that isn’t English-first
50
moving houses and so my home servers are down. Crazy 2 weeks.
πŸ›‘οΈWe added Adobe Commerce vulnerability CVE-2026-48356 & Google Brotli vulnerability CVE-2026-111497 to our KEV Catalog. Visit go.dhs.gov/Z3Q & apply mitigations to protect your org. #Cybersecurity #InfoSec
66
wtf??? Someone please reverse engineer this
i find this pretty funny the M4 iPad Pro with 8GB RAM actually shipped with 12GB RAM, which we found out through teardowns, Apple never acknowledged it and still only let you use 8GB RAM fast forward to 2026, Apple releases a more powerful on device Apple Intelligence model that requires a minimum of 12GB RAM, which won't run on the base model M4 iPad Pro cause it has "8GB RAM", even if the underlying hardware had 12GB RAM make it make sense Apple
1
1
375
gatekeeping models is the new walled garden
anthropic just killed the REA it stops claude from doing it that’s why local models are important
1
95
so fucking annoyed with Railway right now. What a fucking disaster. And the Google Cloud Console is also such a pain to use.
1
107
Manan πŸ€¦πŸ½β€β™‚οΈ retweeted
You’re all prompting wrong
128
587
254
8,302
1,074,395
I’m done with Railway.
At Railway we take security very seriously. Please read this urgent advisory.
126
anytime I've opened twitter today, there's been a cyber clusterfuck somewhere.
82