Anthropic's latest AI model has found more than 500 previously unknown high-severity security flaws in open-source libraries with little to no prompting, the company shared first with Axios.

Why it matters: The advancement signals an inflection point for how AI tools can help cyber defenders, even as AI is also making attacks more dangerous.

Driving the news: Anthropic debuted Claude Opus 4.6, the latest version of its largest AI model, on Thursday.

Before its debut, Anthropic's frontier red team tested Opus 4.6 in a sandboxed environment to see how well it could find bugs in open-source code.
The team gave the Claude model everything it needed to do the job — access to Python and vulnerability analysis tools, including classic debuggers and fuzzers — but no specific instructions or specialized knowledge.

Claude found more than 500 previously unknown zero-day vulnerabilities in open-source code using just its "out-of-the-box" capabilities, and each one was validated by either a member of Anthropic's team or an outside security researcher.

What they're saying: "It's a race between defenders and attackers, and we want to put the tools in the hands of defenders as fast as possible," Logan Graham, head of Anthropic's frontier red team, told Axios.

"The models are extremely good at this, and we expect them to get much better still."

Shorted cyber sec names $CRWD / $ZS @ $400 / $176

Not naked shorts btw, meant to partially hedge out $NET long

Past week has been wild - it’s only February 26, imagine being in 2030. Recapping the key moving pieces:

1) Has capex numbers peaked? If capex hasn’t peaked yet, are hyperscalers going to turn to debt now as forward-looking FCF turns negative with some of these capex projections by GOOG/AMZN?

2) A sentiment shift from “what can AI Do” -> “what can AI not do”. The former was complacency holding on to positions: “it will be ok..”, and the latter is max-fear uncertainty sell-off “get OUT while you can”. This is best reflected in SaaS names. Kinda surprised it took so long for this sentiment gap to converge though. Every software name is now guilty till proven innocent, the entire sector is treated like cancer. Even Blue Owl was talking about their software loan exposure yesterday.

Think much of the SaaS problem is in the initial sector valuation - which was sky-high as sector was thought to be high switching cost, asset light, high incremental margins, high growth etc. now it’s perceived as low/no moat, operating lvg works both ways, low/no growth, customers switching may eventually become as easy as typing in a prompt.

If you look at multiples today (after the sell-off) in the former industry context, then maybe you think it’s oversold. but in the latter context, valuation is probably just decent to maybe still on the higher-end even. So what’s the right industry multiple? I think we eventually settle somewhere on the equilibrium between the two rather than on the extremes, and now I’m leaning neutral on the sector, may become cautiously bullish in next few weeks after the fear has been sufficiently front-loaded. Right now though, I think allocating your mental capacity to finding the winners from the massive productivity unlock from AI > knife-catching the SaaS bottom.

TLDR

https://x.com/eliant_capital/status/2019605852789825856?s=46

Had closed $AG around $26-$27 for 40% gain before the silver sell-off

And just rebought this $SLV correction @ $66.8 in size. Not sure if it’s THE low, but it’s good enough for me. We’ve got to help the uncles queuing at BullionStar out. On a serious note though, long-term structural tailwinds and physical shortage of the metal makes the environment very much in favour of longs still

When the environment is good we tend to spray and pray more. Past week’s mayhem served as a reminder for me to clean up the portfolio and cut low-conviction ideas.

Long exposure is now roughly equal parts: Optics, Memory (NAND only), Agentic AI Software (not to be confused with SaaS), Silver (no miners) and Power (prioritise go-to-market speed). I like this positioning now

In respective order:
$LITE $TSEM
$SNDK $285A $PSTG
$NET $DOCN
$SLV
$BE

Landslide victory by pro-stimulus Sanae Takaichi - largest margin in post-war Japan

Nikkei +5%, precious metals also +5%. Our silver buys last Friday is +10% since entry.

Every major nation is metal to the pedal in running it hot.

Not a fully fleshed-out thought yet.. but this is slowly becoming one of my first principles in investing for the next 5-10 years. Implications extend beyond precious metals.

Looking at $MNDY and $UPWK prints yesterday, I probably closed those too early. Lesson there for myself is to learn to press my high-conviction shorts and even consider doubling down when the thesis is working, rather than being positioned too cautiously for a squeeze that may never come.

This prompted me to take a look at my L/S trading stats over the past 1 year:
- 200+ total tickers traded, with only 40+ names shorted (19%).
- However, these 40+ shorted names generated 33% of my LTM net profit.

While I can learn to be more aggressive in pressing shorts, I'm still quite happy with the overall results! Though this was likely boosted by the high market dispersion over the LTM. I'm much more selective with shorts as I honestly think most people should just not short given that the market is designed to go up. Shorts should mostly be viewed as hedges in my opinion, and rarely an independent speculative position.

$NET is reporting tmr 6am SGT.

I think it will be a beat-and-raise. SW infras names ($DDOG, $DT) have reported numbers that are better than expectations, and Cloudflare’s. channel checks also seem alright with security/zero trust being major growth drivers.

Also Matthew Prince is vibing excessively on Twitter right before earnings.. OGs know what that means..

Risk is that a beat may have already been priced in given DDOG/DT prints & valuation is high. But I like the r/r holding the stock into print and don’t mind buying dips post-print (if it gives) as I think it’s a structural long-term winner.

Also think this earnings season is driving SaaS sentiments to a fork in the road

Investors can now see some Infrastructure SaaS (DDOG, NET, DT) as potential beneficiaries of Agentic AI, while remaining bearish on the other general SaaS that are facing existential threats from Agentic AI.

Perhaps not the entire SaaS sector is destined for the dumps after all

Have closed most of my SaaS shorts, including $CRWD / $ZS at roughly breakeven yesterday. As I said last week was turning neutral on the sector and now I think r/r favours longs. Lots of quality businesses irrationally sold for cheap so it’s time to go dumpster diving

Lots of tidbits in the $NET call, but generally in line with our thesis above

Man this death-to-SaaS thesis might actually be one of my most prescient calls EVER. Turns out being zeroed out overnight was not an exaggeration.

Also interesting that there was only a 1 month lag between my thesis to when the market realized and priced it in; i.e. you only have to look a month ahead these days to make money even on such a large narrative switch — longer than that and your thesis might be far too early given the dynamics of the repricing process (calm for 3 weeks, violent repricing in last week)

Never doubt the Matthew Prince #-of-Tweets Signal

Seems like DeepSeek v4 mini is out, though no official confirmation yet. Context window has increased from 128k to 1m tokens

V4 is rumoured to have a new architectural design relating to how the LLM handles data storage and retrieval. The engram conditional memory module is expected to drive NAND demand as it employs a tiered memory approach that utilises SSD to handle storage overflow from long-context agentic workflows, instead of relying exclusively on expensive HBM/DRAM.

Coincidentally $SNDK is trading at local lows around $545 from investors front-running $WDC share sale… Seems too good to be true but what if?