Anthropic’s Claude Is Pumping Out Vulnerable Code, Cyber Experts Warn

TrustedSec CEO Dave Kennedy says Claude Opus code quality has deteriorated by 47.3% over the past five weeks, with the model introducing serious defects and security issues. Veracode’s testing found Opus 4.7 produced vulnerable code in 52% of 80 coding tasks, versus about 30% for OpenAI models. Anthropic says it is investigating the degradation claims, but the report raises reputational and adoption risks for its AI coding products.

Analysis

The key market read is not that AI coding models are imperfect, but that they may be entering a trust-destruction phase in enterprise workflows. If benchmark quality is slipping while secure-code failure rates remain materially higher than the best-in-class competitor, the adoption curve can flatten exactly where vendors need usage to monetize: daily developer seat expansion and premium inference consumption. That creates a second-order bear case for “AI productivity” narratives because the buyer’s willingness to pay is tied less to raw capability than to error tolerance and governance overhead.

AMD is the most exposed ticker here because the article surfaces a credibility problem from a high-profile customer adjacent to the AI infrastructure stack, not because its product is implicated directly. The more important implication is that enterprises may increasingly cap model usage, shift workloads to smaller/cheaper models, or pull sensitive codegen on-prem, which can reduce demand intensity for frontier-model inferencing and compress the mix toward lower-ARPU deployments. In the near term, that is a sentiment headwind for the AI ecosystem; over months, it could translate into slower growth expectations for compute-heavy beneficiaries if customer ROI audits become more stringent.

AAPL is a relative beneficiary if the market starts pricing a broader “local/privileged AI” migration: device-level inference, private data boundaries, and on-prem control are all aligned with Apple’s privacy brand and installed base. The article also implicitly supports a governance premium for vendors that can prove deterministic controls, auditability, and enterprise security rather than just raw model capability. PLTR benefits at the margin from any re-prioritization toward secure workflows and controlled deployment, though the impact is more narrative than fundamental in the next quarter.

Contrarian take: this may be less a permanent degradation in frontier model quality than a product-optimization tradeoff that temporarily hurts perceived reliability. If the market overreacts, the best expression is not a naked short on AI broad beta but a relative-value trade: short the names most levered to unconstrained AI seat expansion versus long companies selling control, security, and deployment governance. The risk is that vendor fixes restore trust quickly, in which case the dislocation could mean-revert within one product cycle.

Key Decisions for Investors

• Short-term: fade AI infrastructure exuberance with a small AMD short or put spread into the next 2-6 weeks; thesis is not hardware demand collapse, but a near-term multiple reset if enterprise AI ROI scrutiny tightens. Cover if management commentary shows no slowdown in data-center order momentum.
• Pair trade: long AAPL / short RDDT for 1-3 months. If enterprise and consumer users migrate toward private, on-device, or controlled AI experiences, Apple benefits from trust and distribution while RDDT has little direct upside from the theme and no offsetting balance-sheet catalyst.
• Long PLTR on a 2-4 month horizon as a governance/secure-AI beneficiary, but size modestly. Use it as a relative winner versus pure model vendors; upside is multiple support if security-first AI becomes a procurement standard.
• Avoid chasing broad AI beta until vendors demonstrate stable secure-code performance for at least one full release cycle. The risk/reward is unfavorable for outright longs when the key enterprise buyer objection is reliability, not capability.
• If you want convexity, buy medium-dated call spreads on AAPL into the next earnings cycle. The setup is a sentiment-driven re-rating if the market starts rewarding trusted, controlled AI distribution over raw model scaling.