A Leak That Shakes the Markets
On March 26, 2026, Anthropic accidentally made ~3,000 internal documents publicly accessible due to a CMS configuration error. Among them: a draft blog post describing Claude Mythos (internal codename: "Capybara"), a new AI model positioned above the current Opus lineup.
Security researchers Roy Paz and Alexandre Pauwels discovered the leak. Within 48 hours, the market reacted — but not the way you might expect.
The Sell-Off in Numbers
Cybersecurity Stock Crash on March 27, 2026 (%)
The carnage was widespread. In a single session, billions of dollars in market cap evaporated from cybersecurity leaders. CrowdStrike dropped from ~$396 to $369.58 (-5.87%), Palo Alto from $162 to $147, Zscaler extended its decline to -40% YTD.
CrowdStrike (CRWD)
$369.58
Palo Alto (PANW)
$147.07
Zscaler (ZS)
$139.41
Bitcoin
$65,720
Price Trajectory — Before and After the Leak
Cybersecurity Stocks Around the Mythos Leak (base 100 = March 24)
The chart is telling: while the S&P 500 bounced back at month-end (+2.91% on March 30), cyber stocks stayed down. This signals the sell-off isn''t just a risk-off move — it''s a structural repricing of the sector.
Why Cyber Stocks Are Plunging (Not Rising)
The intuitive reaction: "More powerful AI → more threats → more cybersecurity demand → stocks go up." That''s what many retail investors thought. Wall Street saw something else.
The Real Fear: Obsolescence
The leaked draft describes Mythos as "currently far ahead of any other AI model in cyber capabilities." It can discover and exploit software vulnerabilities at speeds substantially faster than human defenders.
Danger
AI provides more "capability uplift" to attackers than to defenders — and that gap is widening. A model like Mythos can elevate an ordinary hacker to nation-state adversary level. Traditional defense tools (signature-based, rule-based) become structurally inadequate.
If AI makes attacks 100x more powerful but defenses only 10x more effective, current cybersecurity products lose their value proposition. That''s the scenario the market priced.
The Stifel Analysis That Hurt
Analyst Adam Borg (Stifel) warned Mythos has "the potential to become the ultimate hacking tool." Evercore questioned whether LLMs could erode cybersecurity vendors'' competitive advantage. Two downgrades in 48 hours — the algos did the rest.
Bitcoin: Collateral Damage
Bitcoin (USD) — Post-Mythos Crash
Bitcoin dropped from $71,300 to $65,720 in 4 days (-7.8%). The Mythos leak wasn''t the only factor ($14.16B options expiry, Iran tensions), but it amplified the broad risk-off move. Crypto, correlated to tech sentiment, got swept in the wave.
The S&P 500 Domino Effect
March 2026 ended at -0.7% for the S&P 500 — 10 of 11 sectors in the red. The March 30 rally (+2.91% to 6,528) masks a month of extreme nervousness. The S&P 500 Equal Weight dropped -6%, revealing the correction is much deeper than the cap-weighted index suggests.
The Debate: Threat or Opportunity?
Claude Mythos — Two Market Readings
Bearish reading (current consensus)
- AI advantages attackers more than defenders
- Traditional cyber tools becoming obsolete
- Zscaler -40% YTD: market prices structural risk
- APTs deploy immediately — enterprises take 6-12 months
- Two analyst downgrades (Stifel, Evercore) in 48h
Bullish reading (contrarian)
- More threats = more cyber budgets (CISOs have no choice)
- BTIG and others call fears overblown
- CrowdStrike: price target $665 (+41% upside)
- Palo Alto: consensus Strong Buy, median target $210
- Historically, every new threat increased cyber spending
Le marché a raison de pricer le risque. Mais à 12-18 mois, les budgets cyber vont augmenter. Le sell-off est une opportunité pour les patients.
Key Takeaway
Attention
The Claude Mythos leak is a wake-up call, not an isolated event. Every AI capability leap accelerates the attacker/defender asymmetry. Cyber stocks that will survive are those integrating AI natively (CrowdStrike Falcon, SentinelOne Purple AI) — not those selling rule-based firewalls from the 2010s. Zscaler at -40% YTD could be a bottom or a trap — wait for Q2 earnings confirmation before adding.
Go from theory to practice
Our Excel templates integrate all the formulas and methodologies presented in this article.
Browse templates