Sam Altman addresses memory remarks, OpenAI patches ChatGPT flaw, meets Samsung and SK Group

Altman’s memory remarks spark debate and mockery

Sam Altman, CEO of OpenAI, addressed criticism on X after temporarily renaming his profile to "Ram Altman." The change followed a joke about memory allocation in home computers, which sparked backlash and mockery, including from Elon Musk. Altman clarified that his earlier remark—suggesting data centers use memory more efficiently than individual devices—was taken out of context but reaffirmed his support for increasing global memory supply.

OpenAI declined to comment on the incident. Memory prices have surged due to demand from AI data centers, with SK Hynix and Micron projecting shortages until 2030 and 2028, respectively. Altman’s comments arrive as OpenAI and peers face capacity constraints for AI infrastructure.

OpenAI fixes ChatGPT security flaw exposing sensitive data

OpenAI patched a server-side vulnerability in ChatGPT’s Lockdown Mode after security researchers at Check Point Research (CPR) demonstrated a bypass technique. The flaw allowed attackers to escape sandbox restrictions, potentially enabling data theft, phishing, malware delivery, and zero-click attacks—where malicious popups appear without user interaction.

The vulnerability exploited how ChatGPT handled HTML previews in code blocks, allowing nested sandboxed frames to bypass restrictions. CPR disclosed the issue to OpenAI, which implemented fixes, but urged organizations to consider the broader attack surface of AI applications beyond the model itself.

Fake AI event invites used to steal corporate credentials

Hackers are targeting corporate credentials by impersonating OpenAI, SpaceX, and Anthropic events. Fraudulent invitations—such as fake livestreams for OpenAI Academy or private SpaceX briefings—promise rewards like free ChatGPT Pro subscriptions or SpaceX shares to lure victims into logging in with Microsoft or Google accounts.

Attackers use phishing-as-a-service platforms Kali365 and VoidProxy to execute these scams. Microsoft credentials trigger device-code phishing, while Google logins route victims through reverse-proxy attacks that bypass multi-factor authentication (MFA). Links are shortened using Google and Zapier to evade security detection, exploiting employees’ interest in AI tools as a high-success phishing lure.

OpenAI balances collaboration and competition with European startups

OpenAI is positioning itself as a "catalyst" for European startups, despite tensions over competition. The company collaborates with startups like Swedish legaltech Legora, which co-developed a plugin with OpenAI, even as OpenAI launches rival products such as Astra for Law. Tricia Troth, OpenAI’s head of go-to-market for startups in EMEA, emphasized the company’s role in supporting startups from pre-funding to scaleup stages, including technical, engineering, and marketing assistance.

OpenAI’s startup team—comprising former founders and venture capitalists—also connects startups with investors and developers. However, European AI application-layer startups increasingly compete with OpenAI and Anthropic, raising concerns about market consolidation.

Altman seeks memory chip partnerships amid AI infrastructure shortages

Sam Altman met with executives from Samsung and SK Group in Korea to discuss AI technology development, including potential cooperation on memory chips. While Samsung described the meeting as a seminar on AI’s impact on management, investors and analysts interpreted it as an effort to secure supply amid global shortages. No new agreements were announced, and both Samsung and SK hynix denied discussions on chip pricing or investment commitments.

The timing underscores the strain on AI infrastructure capacity. Samsung’s 2027 high-bandwidth memory (HBM) supply is sold out, and SK hynix’s DRAM and HBM output is booked through 2029. Altman described the current environment as "very supply-constrained" and predicted memory manufacturers could generate $100 billion in annual free cash flow. Samsung is shifting toward long-term supply agreements with price floors and advance payments.

Broader industry context

The bypass of ChatGPT’s Lockdown Mode highlights risks in AI application layers beyond the model itself. Earlier reporting noted user backlash over OpenAI’s merged usage limits for ChatGPT and Codex, reflecting broader concerns about transparency and control in AI tools. Meanwhile, global protests against data centers and infrastructure shortages—such as Texas’s proposed permit freeze—mirror the capacity challenges Altman referenced in his memory remarks.

Anthropic’s recent loss of product lead Catherine Wu to OpenAI, alongside other departures ahead of its IPO, also underscores competitive pressures in the AI sector.

Related on Lazyfounder

• OpenAI
• Texas data center freeze and global protests split AI infrastructure sector
• Anthropic’s Claude Code product lead Catherine Wu joins OpenAI
• OpenAI’s new shared usage limits spark backlash and questions
• More technology news