Lumina Digest

Micron Crosses $1 Trillion Market Cap, Driven by the AI Memory Shortage

On May 26, 2026, Micron joined the trillion-dollar club after a 19% jump in a single session, propelled by demand for high-bandwidth AI memory and a record upgrade from UBS.

Micron Technology topped a $1 trillion market capitalization for the first time on May 26, 2026, after a jump of roughly 19% in a single session — its best daily gain since 2011 — with the stock closing around $895. The Boise-based manufacturer thus becomes the eleventh-largest U.S. publicly traded company, overtaking Walmart in market value after more than tripling its share price since the start of the year.

The immediate catalyst was a record upgrade: UBS analyst Timothy Arcuri tripled his price target from $535 to $1,625, the highest on Wall Street. But the underlying driver is structural and comes from high-bandwidth memory (HBM), an essential component for AI training and inference. Micron controls about 21% of the HBM market and, together with the only two other producers at scale — Samsung and SK Hynix — has already sold out its HBM capacity for all of 2026, managing to cover only 50-65% of customers' medium-term demand.

The scarcity is feeding through to prices: according to Counterpoint Research, memory prices have risen 80-90% since the start of the year. Production expansions in Singapore and Taiwan will not add capacity before the end of 2027, and again according to Counterpoint Research, supply relief will not arrive before the fourth quarter of 2027. As reported by gagadget and Crypto Briefing, the market is recognizing that the AI constraint is not just the GPU, but also the memory that feeds it.

Why it matters

• Entrepreneurs · ICT engineers / IT managers: With HBM capacity already sold out for 2026 and DRAM under heavy supply pressure, and with memory prices up 80-90%, memory procurement becomes a direct variable in cost, planning, and resilience. Anyone buying servers, GPUs, or AI capacity must factor in multi-year contracts, lock-in with the few suppliers (Micron, Samsung, SK Hynix), and a shortage that will not ease before 2027-2028. For those managing IT roadmaps and budgets, it means placing orders early and assessing the impact of rising memory costs on the TCO of AI infrastructure.

Drupal Core, Critical SQL Injection (CVE-2026-9082): Active Exploitation and Patch Mandated by CISA

An unauthenticated SQL injection flaw in Drupal core is already under mass attack. It affects only sites running PostgreSQL, but CISA has added it to the KEV catalog with a federal patch deadline of May 27.

Drupal's security team has published advisory SA-CORE-2026-004 for CVE-2026-9082, a SQL injection vulnerability in the core's database abstraction API. The flaw lies in the EntityQuery condition handler for PostgreSQL: user-controlled PHP array keys reached SQL placeholder construction without sanitization. The fix applies array_values() to discard attacker-supplied keys. The flaw is exploitable remotely and without authentication, with possible information disclosure, privilege escalation and, in some cases, code execution. The discovery is credited to Michael Maturi, a researcher at Google/Mandiant.

One decisive detail about the scope: only sites with a PostgreSQL backend are vulnerable; MySQL, MariaDB and SQLite are not. Tenable still reports a CVSS of 6.5, but NVD/Drupal CNA now indicates CVSS 9.8 ("Critical"). Drupal's internal risk score was 20/25 in the initial advisory and was updated to 23/25 on May 22, after exploit attempts in the wild ("Highly Critical"). The fixed versions include 11.3.10, 11.2.12, 11.1.10, 10.6.9, 10.5.10 and 10.4.10.

The patch was released on May 20; the advisory, initially classified as a theoretical risk, was updated on May 22 when exploit attempts appeared in the wild. Imperva observed more than 15,000 attack attempts against nearly 6,000 sites in 65 countries in the first two days, concentrated on gaming and financial services (~50%). For now the attacks are predominantly in a reconnaissance and validation phase. CISA has added the flaw to the KEV catalog, ordering federal agencies (FCEB) to apply the patch by May 27, 2026.

Why it matters

• ICT engineers / IT managers: It calls for immediate emergency action: update every Drupal installation on PostgreSQL to the fixed versions right away, because exploitation is massive and indiscriminate (nearly 6,000 sites targeted) and the CISA deadline is a compliance reference even outside the federal perimeter. Above all, remember to map the database backend as a risk parameter in its own right: the same application is exposed or safe depending on whether it runs on PostgreSQL or on MySQL/MariaDB, so the asset inventory must record the DB engine, not just the app version.

Ghost CMS: a SQL Injection in the Content API Bends Over 700 Sites Into ClickFix Campaigns

The CVE-2026-26980 flaw (CVSS 9.4), patched in February but left unpatched on hundreds of installations, allowed attackers to steal Admin API Keys and turn trusted sites — Harvard, Oxford, DuckDuckGo — into traps that distribute malware to their readers.

A critical vulnerability in Ghost CMS, the publishing platform used by more than 100,000 sites, has been mass-exploited to compromise over 700 domains and use them as malware distribution vectors. The flaw, tracked as CVE-2026-26980 and rated CVSS 9.4 by GitHub/CNA (NVD instead assigns it 7.5), is a SQL injection in the Content API: an unauthenticated attacker can read any database table — including Admin API Keys, bcrypt password hashes, and session secrets — by injecting payloads into the filter=slug or order=slug parameters. It affects versions from 3.24.0 through 6.19.0; the 6.19.1 patch was released by Ghost on February 16, 2026, while the CVE was published on NVD on February 19, 2026.

The campaign, detected in early May, follows a ClickFix chain: once the Admin API Key is stolen, the attackers use the legitimate API to bulk-edit articles and inject JavaScript loaders at the bottom of the page. Visitors to the compromised sites are shown a fake Cloudflare verification page that prompts them to paste a command into the Windows prompt. The command downloads payloads such as a DLL loader, a JavaScript dropper, and an Electron executable named UtilifySetup.exe. Victims include Harvard University, Oxford University, Auburn University, and DuckDuckGo. According to the analysis by XLab (QiAnXin), at least two distinct groups operate in parallel, competing for the same infected sites; about half of the victims are personal blogs, the rest are software, AI, and cryptocurrency outfits. Remediation requires upgrading to at least 6.19.1, rotating all keys and credentials, and removing the scripts injected at the database level.

Why it matters

• ICT engineers / IT managers: A patch available since February is not enough if it is not applied: upgrading to Ghost ≥6.19.1 is necessary but not sufficient, because the Admin API Keys may have already been exfiltrated. You need to rotate all keys and credentials, verify the integrity of published pages, and clean up the scripts injected directly into the database.
• End users: The risk comes from sites you consider trustworthy: a page from a university or a well-known service can display a fake Cloudflare verification. The defensive rule is clear-cut — no legitimate site ever asks you to paste and run commands in the prompt or terminal to "prove you're not a robot".

GitHub Rotates GitHub Enterprise Server Signing Key After Unauthorized Access to Internal Repositories

A poisoned VS Code extension (Nx Console) compromised an employee's device, exposing roughly 3,800 internal repositories. GitHub is requiring self-hosted GHES administrators to rotate the signing key.

GitHub stated that it detected and contained the compromise of an employee's device on May 18, 2026, caused by a poisoned version of the VS Code Nx Console extension. According to GovInfoSecurity's coverage, the malicious extension remained published on the Visual Studio Marketplace for just 18 minutes before being removed, but that was enough to trigger the incident.

The attacker claimed access to roughly 3,800 repositories, a figure GitHub describes as "directionally consistent" with its own investigation. Only GitHub's internal repositories were affected, some of which contained "excerpts of support interactions"; the company says it has "no evidence" of impact on customer data. GitHub.com and GitHub Enterprise Cloud are not involved.

The operational burden, by contrast, falls on administrators of self-hosted GitHub Enterprise Server (GHES). CISO Alexis Wales asked that all keys be rotated, including the GHES signing key. Administrators must run an official script (SHA256 digest 3009bf5c…6b43) that regenerates the GPG public key on both the admin account and root. The procedures differ for single-node topology and for cluster/HA (ghe-cluster-each). Future patches will be signed with the new key: without rotation, upgrades will fail with the error "The file provided is not a valid GitHub Enterprise Server package".

Why it matters

• ICT engineers / IT managers: Anyone managing self-hosted GHES instances must act now: rotate the signing key by following the official script (verifying its SHA256 digest) before the next round of updates, otherwise patches will be rejected. The incident is also a supply-chain reminder: third-party IDE extensions are a real vector for compromising developer endpoints, and they deserve approval and monitoring policies on par with any other dependency.

Pope Leo XIV Dedicates His First Encyclical to Artificial Intelligence and Condemns AI in Warfare

In his first encyclical, Magnifica humanitas, Pope Leo XIV places the human person at the center of the AI era. Technology «is never neutral», artificial intelligence cannot hold supremacy over humankind, and its use in war must be condemned.

Pope Leo XIV has presented his first encyclical, Magnifica humanitas (about 42,300 words), devoted to safeguarding the human person in the age of artificial intelligence. The document is dated May 15, 2026, but the Pontiff presented it in person at the Vatican on May 25: an unusual gesture, normally delegated to the cardinals. In attendance was an audience of industry experts, including Anthropic co-founder Chris Olah.

The text does not portray technology as humankind's enemy, nor as intrinsically evil. It warns, however, that it «is never neutral, because it takes on the characteristics of those who design it, fund it, regulate it, and use it». The underlying choice is conveyed through two biblical images: building a new Tower of Babel, or raising up the city in which God and humankind dwell together. AI, devoid of experiences, values, and feelings, cannot and must not take on a role of responsibility or of supremacy over human intelligence. Otherwise, the risk is a «less human» civilization and the devaluation of work.

As Vatican News summarizes, the encyclical denounces how AI amplifies the power of those who already command economic resources, expertise, and access to data. It then argues that data ownership cannot remain in private hands alone, and it calls for adequate regulatory tools. It also explicitly condemns the use of AI in war: reduced human control over weapons makes it easier to start conflicts. It even goes so far as to call the traditional doctrine of just war «outdated». It is the first encyclical published without an official Latin version, and it follows in the tradition of the social doctrine begun by Leo XIII. The stance was picked up internationally as the first major magisterial text devoted entirely to AI.

Why it matters

• End users: It offers hundreds of millions of people a shared moral framework for deciding how much to delegate to automated systems: it urges them not to give in to over-reliance and to safeguard human dignity and relationships wherever AI threatens to replace them.
• Entrepreneurs: The appeal to «never neutral» technology, to the concentration of power and data, and to the value of human labor sets ethical and reputational expectations — and foreshadows regulatory pressure — on how to design, govern, and communicate AI adoption within the company.

7-Eleven Data Breach: 185,000 People Affected and ShinyHunters' Extortion

7-Eleven is notifying over 185,000 people after a ShinyHunters intrusion that began with a server holding franchisee documents; the group claims over 600,000 Salesforce records and an extortion attempt.

7-Eleven has confirmed a data breach that exposed the personal data of over 185,000 people. According to chief information security officer Jim Kastle, on April 8, 2026 the attackers breached an internal server containing franchisee documents. These are the forms submitted during the application process to open a store (TechCrunch).

The compromised data includes names, dates of birth, addresses, phone numbers and emails. For a subset of those affected, U.S. Social Security numbers and driver's license numbers were also exposed, as documented in the notifications to the Massachusetts attorney general's office. Have I Been Pwned puts the number of exposed accounts at roughly 185,300; the Maine filing reports 2 Maine residents and does not state an overall total (SecurityWeek).

The attack is claimed by ShinyHunters, a group specializing in hack-and-extortion that since mid-2025 has been targeting the Salesforce instances of large organizations. Organizations previously linked to or claimed by the group include Google, Cisco and the European Commission. The claim involving Cisco is not publicly confirmed. The group says it stole over 600,000 Salesforce records containing PII and internal corporate data. It had set April 21 as the ransom deadline, threatening to publish the data (SecurityAffairs). A discrepancy remains between the ~600,000 records claimed by the attackers and the ~185,000 individuals confirmed as affected so far. With the payment not made, the data was released online and indexed by Have I Been Pwned.

Why it matters

• Entrepreneurs: The case shows the real bottom-line cost of a breach: regulatory notifications across multiple states, exposure to litigation and extortion demands, reputational damage. And the incident starts with the data of applicants and franchisees, not customers: the information of anyone proposing a partnership or a supply arrangement is also an attack surface and a compliance risk.
• ICT engineers / IT managers: 7-Eleven confirms access to systems holding franchisee documents; ShinyHunters claims, without confirmation from the company, that Salesforce was involved. The data of applicants and suppliers must be segmented, minimized and protected like that of customers. ShinyHunters' modus operandi against SaaS platforms confirms that the security posture must extend to third-party systems and to the 'secondary' datasets that are often overlooked.

OpenAI Releases Workload Identity Federation and Opens a Secure MCP Tunnel for the Enterprise

On OpenAI's developer platform, trusted workloads exchange OIDC tokens for ephemeral credentials, with no more static API keys. An egress-only tunnel connects private MCP servers without exposing them to the Internet.

On May 26, 2026, OpenAI released workload identity federation on its developer platform. According to the changelog, trusted workloads can "exchange externally issued identity tokens for short-lived OpenAI access tokens without storing long-lived API keys". In practice, a service exchanges an OIDC token issued by its own infrastructure for a short-lived OpenAI access token. This removes static API keys from production systems. The supported identity providers are Kubernetes, AWS, Microsoft Azure, Google Cloud, and GitHub Actions.

The official guide describes four components. A workload identity provider defines the trusted OIDC issuer (URL, audience, key source). A mapping authorizes specific token attributes to generate credentials for a given service account. Then come the token exchange and the use of the OpenAI token as a bearer for API calls. Attribute transformations use Common Expression Language (CEL); configuration takes place in Organization Settings > Security, and Admin API scopes cannot be assigned to mappings.

A few days earlier, on May 19, OpenAI had introduced the Secure MCP Tunnel for enterprise customers. The tunnel lets ChatGPT web, Codex, the Responses API, and AgentKit reach private or on-prem MCP servers through a tunnel-client hosted by the customer. The link is an egress-only HTTPS connection. The tunnel-client long-polls for queued work and forwards it to the local server, so that — as the guide states — "the private MCP server does not need a public listener". No open inbound ports are required; optional mTLS, outbound proxy, and custom CA bundles are supported.

Why it matters

• LLM builders / devs: For those building agents and services, ephemeral credentials tied to the workload's identity remove from code and deployments the static API keys that have to be stored and rotated. The Secure MCP Tunnel also makes it possible to connect private or on-prem MCP servers to agents without publishing them on the Internet.
• ICT engineers / IT managers: OIDC federation hooks into the infrastructure already in use (Kubernetes, AWS, Azure, GCP, GitHub Actions) and moves secret control to the organization level, reducing the attack surface of long-lived credentials. The egress-only tunnel avoids opening inbound ports to integrate internal systems, preserving the network perimeter.