BlockBeats reported on July 25th that Citrini analyst Jukan responded to recent market rumors regarding bearish NAND notes and negative news about QLC price negotiations. He stated that it's not surprising that SanDisk accepted a lower price than initially quoted when signing a long-term agreement with Meta. SanDisk is one of the most aggressive NAND manufacturers in vying for LTAs, planning to allocate over 50% of its total shipments to such agreements. Based on this strategy, it's natural that they are willing to accept LTA prices lower than the current quarterly contract price. Therefore, it cannot be inferred that "SanDisk cannot seamlessly resell all orders to North American customers who offer higher prices."

Regarding rumors that Chinese module manufacturers were rejected when trying to sell eSSDs to domestic CSPs, Jukan explained that Chinese CSPs have direct procurement channels from Yangtze Memory Technologies Co., Ltd. (YMTC), rather than a lack of demand. As for the claim that hyperscale cloud providers are pressuring for lower prices on QLC eSSDs, resulting in unsold volumes, he believes that new cloud providers have sufficient demand to absorb these volumes.

Jukan concluded by saying that negative headlines are more easily amplified when memory stocks underperform, but the fundamentals have not deteriorated substantially. He reiterated his bullish stance on memory. Previously, Jukan had stated that DRAM contract prices still had about 40% upside potential by the end of 2027 and that HBM supply remained tight. This clarification regarding NAND further solidifies his bullish position on the entire memory sector.

According to Beating's monitoring, Anthropic disclosed in its Claude Opus 5 system card that its model is virtually immune to cue injection attacks in browser agent scenarios, remaining unbroken in 129 test scenarios. This breakthrough is significant, as OpenAI publicly acknowledged last December that cue injection may never be completely eradicated. In Gray Swan's general cue injection benchmark test, Opus 5's success rate after 15 attacks was only 2.0%, significantly better than the previous generation Opus 4.8's 5.5%, and also ahead of Mythos 5's 2.6% and Fable 5's 2.8%.

Hint injection is considered the most serious security vulnerability facing AI agents—attackers can bypass model instructions and execute malicious operations by embedding manipulated input content such as hidden text in web pages. A zero attack rate was only achieved in products like Claude Cowork with Auto Mode enabled. This achievement marks a shift in AI agent security from "continuous patching" to "structural defense."

Click the original link below to join the Beating · Lark AI news channel and monitor global AI hot topics and news 24/7.