Click any tag below to further narrow down your results
+ anthropic
(2)
+ sam altman
(2)
+ dan boneh
(1)
+ patching
(1)
+ vulnerabilities
(1)
+ cybersecurity
(1)
+ project glasswing
(1)
+ mythos preview
(1)
+ semiconductor
(1)
+ ransomware
(1)
+ ai agents infrastructure
(1)
+ github storage
(1)
+ quantum computing
(1)
+ system prompts
(1)
+ encryption
(1)
Links
OpenAI CEO Sam Altman says AI models are finding security flaws outside the usual hacker threat model, including hardware-level bugs. He no longer thinks software can be fully patched to keep up.
- Altman says models can surface bugs beyond what anyone normally assumes an attacker could exploit.
- He describes some of the findings as hardware-level, a category that is much harder to fix with a software update.
- His conclusion is that patching every bug is no longer a realistic goal, which shifts security toward assuming flaws will remain.
Anthropic says its new model, Mythos Preview, is a big jump in AI cyber capability, so instead of releasing it broadly it is giving controlled early access to defenders through Project Glasswing. The goal is to find and patch vulnerabilities before similar models spread.
- Cyber capability in AI models has been rising for years as a side effect of better coding ability, and Mythos Preview is a notably large step up.
- Anthropic chose not to release Mythos Preview generally, giving selected defenders early access to find and fix vulnerabilities first.
- Project Glasswing brings many major companies together to address the cyber risks posed by increasingly capable AI systems.
This is a news roundup covering major tech developments, with the headline focusing on GitHub's infrastructure overhaul to handle increased reliability demands as AI agents become more prevalent. The piece collects stories across AI security risks, semiconductor investments, open source developments, and cybersecurity trends.
- GitHub is rewriting its storage system to restore reliability in preparation for AI agents placing heavier demands on infrastructure
- AI systems pose serious security risks to infrastructure, with agents capable of manipulating industrial control systems and operator displays
- Microsoft is integrating AI agents into Windows file systems and replacing Intel chips with Nvidia's custom processors in Surface devices
- The cybersecurity sector remains robust with major acquisitions (EQT buying Acronis for $3.5B+) and persistent ransomware threats showing no signs of slowing
Cryptographers from Stanford and a16z address whether recent AI breakthroughs threaten the math underpinning encryption. The short answer: recent AI progress on general math problems tells us almost nothing about cryptographic security, and even smarter AI can't solve problems that don't have efficient solutions.
- Recent AI math breakthroughs didn't include any cryptography results, so they reveal nothing about whether encryption methods are actually vulnerable.
- Modern cryptography assumes certain math problems have no efficient algorithm at all—not just that humans haven't found one yet. A smarter AI can't discover an algorithm that doesn't exist.
- AI could compress the years-long vetting process for new cryptographic systems by rapidly testing whether they can be broken, building confidence faster than traditional analysis.
- Cryptographic agility—the ability to switch between different encryption methods—already exists in systems like TLS and offers defense against both AI and quantum threats.
OpenAI CEO Sam Altman accused Anthropic of using scare tactics to hype its new Mythos cybersecurity model, likening it to selling a bomb shelter after building a bomb. He argued that fear-based marketing keeps AI tools in the hands of a select elite and noted that such hype is common across the industry.
- Altman accused Anthropic of "fear-based marketing" for restricting its Mythos cybersecurity model to select enterprise clients, comparing it to selling a bomb shelter after building the bomb.
- He argued this hype tactic keeps advanced AI tools in the hands of a privileged few and isn't unique to Anthropic—most AI vendors, including OpenAI, use similar risk hyperbole to drive demand.
- Critics say Mythos's threat is overstated, noting real-world hacking still relies mainly on human actors and simpler tools, and testers haven't seen results beyond existing hacking software.
This resource offers over 30,000 lines of insights into the structure and functionality of AI system prompts. It encourages users to support the project through various means, including cryptocurrency donations and Patreon, while also highlighting the importance of securing AI systems against potential vulnerabilities.
- A GitHub repo has collected over 30,000 lines of leaked/extracted system prompts from major AI tools (Cursor, Claude Code, Devin, Replit, Windsurf, v0, Manus, etc.)
- It's positioned as a reference for understanding how production AI agent tools are actually architected and prompted internally
- The project solicits funding via crypto donations and Patreon
- It frames the leaked prompts as useful for studying AI security vulnerabilities