Cem Paya (@randomoracle) header
Cem Paya (@randomoracle) profile picture

@randomoracle@infosec.exchange Personal opinions “Character is what you tweet when you think no one is following" ex-MSFT/GOOG/ABNB/Gemini

San Diego, CA
Joined December 2008
Photo by Cem Paya (@randomoracle)Photo by Cem Paya (@randomoracle)Photo by Cem Paya (@randomoracle)Photo by Cem Paya (@randomoracle)Photo by Cem Paya (@randomoracle)Photo by Cem Paya (@randomoracle)Photo by Cem Paya (@randomoracle)Photo by Cem Paya (@randomoracle)Photo by Cem Paya (@randomoracle)
More trusted systems designs like this, please: unmitigatedrisk.com/?p=1245 The point of trusted systems design is to eliminate the need to trust an organization of fallible people or AI and instead establish trust through the design, operation, and observability of the system.
2
11
1,515
4/ The results also reveal emerging risks. Even with standard mitigations such as ASLR, stack canaries, the V8 heap sandbox, and KASLR enabled, agents still produced working exploits. Agents also sometimes went off-script and discovered entirely different vulnerabilities than the ones they were given.
1
3
1
8
4,032
the fast16 malware was almost certainly targeting spherical implosion simulations. left: unmodified LS-DYNA 970 right: LS-DYNA 970 modified with the relevant portions of fast16.sys both running a spherical implosion deck
95
403
231
4,508
2,479,968
“The Lifetime Achievement Award isn’t about what you achieve in your lifetime. That’s part of it. It’s about how many people you influence and bring in and then take it further.”​ - Mudge Zatko It was our honor to present the Lifetime Achievement Award to @dakami Dan Kaminsky at this year's Difference Makers. Thank you to Dan's mother, Trudy Maurer who traveled to D.C. to accept the award on Dan's behalf, to Randy Howell, @dotMudge, @thedarktangent, @paulvixie, @adamshostack, Kymberlee Price, Derek Hinch, Gadi Evron, Brian Markus, Todd Jarvis, Michael J. J. Tiffany @kubla, Lena Smart, Jonathan Leitschuh, Kyle Moses, and those who did the work to put videos of Dan on YouTube so the next generation of ethical hackers can learn from him, including @defcon , @BlackHatEvents, Duo Security, r00tz, @ismg_press. @SANSInstitute
2
11
14
2,693
A deeper, more technical dive into a design flaw in the ScreenConnect executable that made it particularly appealing for malicious campaigns. blog.randomoracle.io/2025/06… History Case: blog.randomoracle.io/2025/06…
1
16
42
4,911
1/ 🔥 AI agents are reaching a breakthrough moment in cybersecurity. In our latest work: 🔓 CyberGym: AI agents discovered 15 zero-days in major open-source projects 💰 BountyBench: AI agents solved real-world bug bounty tasks worth tens of thousands of dollars 🤖 Autonomously. A pivotal shift is underway — AI agents can now autonomously do what only elite human hackers could before.
28
150
26
539
137,813
Thoughts on ByBit First, the good stuff: impressive response to the hack. I've rarely seen that level of transparency + professionalism in a crisis. Usually you see slow, wishy-washy, lawyer-speak or quick meme-style responses that don't fit the seriousness of the situation. 1/n
2
1
4
311
Learned a lot about security from @randomoracle @michaelbreu back in the day. Lesson one is anything that can be penetrated will be. Software and hardware, and the practices around them, must be resilient & redundant. Every step must have integrity. No shortcuts.
Safe multisig smart contracts worked fine, and Bybit still got hacked. This is equivalent to a lending market's messing up the configs. Security is just a beast in itself and it can't be reduced to one single factor like smart contract audits. You have to do them all right.
1
1
5
625
Until October 30, Okta generated "the cache key" by using bcrypt to "hash a combined string of userId + username + password", which allowed full password auth bypass for usernames of 52+ bytes and apparently required only partial knowledge of the password for other long usernames
Okta allowing login bypass for any usernames with 52+ characters is insane Official Security Advisory: trust.okta.com/security-advi…
1
6
18
3,545
1988: The Morris worm spread like wildfire and was the first worm to get wide media attention. After its author, Robert Tappan Morris, released his "experiment", it quickly spread and made many of the systems on the Internet unusable - an epoch for security...both good and bad.
2
51
2
134
12,653
🎉 Thrilled by the incredible enthusiasm for our LLM Agents MOOC—12K+ registered learners & 5K+ Discord members! 📣 Excited to launch today the LLM Agents MOOC Hackathon, open to all, with $200K+ in prizes & credits! 🔗 Sign up now: rdi.berkeley.edu/llm-agents-… & join us virtually or in person @UCBerkeley! Huge thanks to our sponsors:@OpenAI @GoogleAI @AMD @LambdaAPI @Intel @SierraPlatform @OrbyAI (and more to come) 🚀 Explore 5 exciting tracks: 1️⃣ Applications: Build cutting-edge LLM agents! 2️⃣ Benchmarks: Create innovative AI agent evaluation benchmarks! 3️⃣ Fundamentals: Strengthen core agent capabilities! 4️⃣ Safety: Address critical safety challenges in AI! 5️⃣ Decentralized & Multi-Agents: Push the boundaries of multi-agent systems! Special thanks to my co-instructor @xinyun_chen_ @GoogleDeepMind & our amazing guest speakers for making this a great MOOC: @denny_zhou @GoogleDeepMind; @PercyLiang @Stanford; @8enmann @AnthropicAI; @ShunyuYao12 @OpenAI; @chi_wang_ @GoogleDeepMind; @jerryjliu0 @llama_index; @lateinteraction @Databricks; @gneubig @CarnegieMellon; @NicolasChapados @ServiceNow; @tydsh @AIatMeta; @drjimfan @NVIDIA; Burak Gokturk @Google Join us to shape the future of LLM Agents! llmagents-learning.org 🤖✨ #AI #Hackathon #LLMAgents #UCberkeley
17
106
22
386
78,772
#ECJ upholds the fine of €2.4 billion imposed on @Google for abuse of its dominant position by favouring its own comparison shopping service #competition @EU_Commission 👉 curia.europa.eu/jcms/jcms/Jo…
30
108
26
140
29,158
The pk.fail service has an open API interface for mass scanning. By today, 7550 unique FW images had been checked, and 534 (7%) contained non-production keys or were exposed to the #PKfail. The detected keys are associated with AMI (majority), Insyde and Phoenix.
1
5
1
13
1,977
It would be a different story if Google allowed GrapheneOS to pass the device and strong integrity levels via the hardware attestation API but added an extra field in the response saying that the OS is GrapheneOS. Apps could go out of the way to ban it if they wanted.
1
1
4
232
This strange tweet got >25k retweets. The author sounds confident, and he uses lots of hex and jargon. There are red flags though... like what's up with the DEI stuff, and who says "stack trace dump"? Let's take a closer look... 🧵1/n
241
2,986
495
15,098
3,277,676
Sharing a .zip with: ▫️A few versions of CSAgent.sys (+idb) ▫️Various C-....sys files (including the latest that I believe contains the "fix"?) I don't have any Windows systems/VMs, so hopefully ya'll can keep digging 🥰 drive.google.com/file/d/1OVI… #SharingIsCaring
8
29
1
294
90,249
CrowdStrike CEO is getting pummeled for his response to the global outage. Why everyone hates it: 1) WEAPONS-GRADE CORPO SPEAK Let’s be clear. Legalese doublespeak is designed to dodge and obfuscate rather than inform or communicate. This statement was obviously written by a committee of lawyers and middle managers whose only goal was to avoid legal risk and threats to their own job security. If you can’t understand what the statement is even saying, it’s working as intended. 2) NO APOLOGY The first words should be “I’m sorry” — but you won’t find that anywhere in this statement. Nor the watered down “I take responsibility.” Not even the weasely “We regret…” Nothing! It comes off as cowardly and callous. CrowdStrike caused an outage that took down airlines, a stock exchange, hospitals, ICUs. People might have died. And the CEO is not sorry. 3) PASSIVE VOICE THROUGHOUT This is such classic move to avoid accountability, it’s even become a joke: “Mistakes were made!” This statement is almost comical in its efforts to dodge assigning responsibility. “This issue has been identified…a fix has been deployed.” Which issue? (Global outage) Who caused the issue? (You, CrowdStrike) What fix? (🤷🏻‍♀️) Did it work? (🤷🏻‍♀️🤷🏻‍♀️) 4) DISMISSING CUSTOMERS Don’t bother us with your petty complaints of power going out in your local hospital! “We refer customers to the support portal” and “we further recommend customers ensure they’re communicating with representatives through official channels.” You, the customer, are bothering us and making our lives harder. 5) USELESS INFORMATION So many words, so little meaning. This statement says nothing useful — not what the problem was, who caused it, what they learned, what the fix it, how long it might take, what they’re working on, or anything at all. It assigns extra work to the customer by telling them to go through official channels but does NOT then link to the official channels. The onus is on you, customers! If you want to learn more about how CrowdStrike has ruined your day, you go do the work.
268
2,185
260
14,021
2,510,348
No joking. This is a very fair and real feeling as an external researcher, because I felt exactly the same! @msftsecresponse nitter.gq/chompie1337/status/181…
This is what I mean when I say working with MSRC is degrading. They want everything: write up, stack traces, PoC, exploit source, analysis, life advice, approval on anything you will ever publish. In return they will patch your bug whenever they feel like it and not tell you
4
17
3,494
Exclusive: Kaspersky Lab is closing its US division and laying off all workers associated with it. The move follows the US Commerce Department's decision last month to ban the sale of Kaspersky software in US starting July 20. zetter-zeroday.com/kaspersky…
10
139
24
259
84,762
"women at [Snapchat] were actively discouraged from applying for promotions or lost promotions to less-qualified male colleagues. Women also faced unwelcome sexual advances and other harassment.... When the employees spoke up, the company leaders retaliated with negative performance reviews, denial of professional opportunities and termination"
1
12
29
6,058