Search: last-key.com
32 threats
Most recently published first
| Severity | CVSS | Signals | Threat | Source | Published | Score and star |
|---|---|---|---|---|---|---|
| High | 8.3 | EPSS 0.4% | CVE-2026-77685zephyrproject zephyr · Bounds High8.3EPSS 0.4%CVE Database V5CVE-2026-77685 The RFCOMM multiplexer control-channel handler in subsys/bluetooth/host/classic/rfcomm.c computed the echo length for a BT_RFCOMM_TEST command as buf->len - 1 and passed it to rfcomm_send_test(), whose length parameter is a uint8_t. A malformed but FCS-valid UIH frame on DLCI 0 can leave buf->len == 0 after rfcomm_handle_msg() pulls the two-byte MCC header, so buf->len - 1 underflows and narrows to 255. rfcomm_send_test() then copies 255 bytes out of a receive buffer that holds no remaining valid data, and appends them to an outgoing L2CAP PDU allocated with only CONFIG_BT_L2CAP_TX_MTU (253 by default for CONFIG_BT_CLASSIC) bytes of tailroom. The source read runs past the end of the ACL receive buffer into adjacent buffers of the same net_buf pool, and the destination copy overruns the transmit buffer by about eight bytes; net_buf_add_mem() performs no runtime bounds check in ordinary builds, where the underlying __ASSERT_NO_MSG() is compiled out. The trigger is a single five-byte RFCOMM frame on the multiplexer control channel, which is processed as soon as the L2CAP channel for PSM 3 is established — before any DLC exists and independent of RFCOMM session state. bt_rfcomm_init() registers that L2CAP server with .sec_level = BT_SECURITY_L1, which l2cap_br_conn_security() accepts without any link key from a peer that does not claim host Secure Simple Pairing support, so an unpaired BR/EDR device in radio range of a connectable target with CONFIG_BT_RFCOMM enabled (SPP, HFP, HSP, GOEP and similar profiles all enable it) can reach the defect. The resulting 255-byte TEST response is transmitted back to the attacker, disclosing roughly 255 bytes of adjacent stack/pool memory — including the contents of neighbouring receive buffers, which an attacker can prime with earlier packets — and the accompanying out-of-bounds write corrupts memory following the transmit buffer, which can crash the Bluetooth host. The request is trivially repeatable, so both the disclosure and the corruption can be driven at will. Two fixes are needed, and the first on its own does not close the out-of-bounds write. Fix 1 (0040ef781e1d, #115768) strips the FCS byte in rfcomm_recv() and echoes buf->len, rejecting anything over UINT8_MAX, which removes the underflow and the over-read. But rfcomm_make_uih_msg() still allocates the response with bt_l2cap_create_pdu() and never compares the requested length with the buffer's tailroom, so a well-formed TEST command carrying a pattern of about 248 bytes or more still builds a response (3-byte frame header, 2-byte message header, pattern, FCS) larger than the 253 bytes available. A pattern that long can only be received when the RFCOMM L2CAP receive MTU (CONFIG_BT_RFCOMM_L2CAP_MTU, which defaults to CONFIG_BT_BUF_ACL_RX_SIZE) exceeds CONFIG_BT_L2CAP_TX_MTU. The plain BR/EDR defaults (200 against 253) cannot reach it, but from v4.3.0 enabling CONFIG_BT_GOEP raises CONFIG_BT_BUF_ACL_RX_SIZE to 264 (265 from v4.4.0), so GOEP builds with only fix 1 applied still overrun the transmit buffer by up to eight bytes. This includes v4.3.2, the last v4.3 release, which carries fix 1 but will not receive fix 2. Fix 2 (214530895dd3, #118293) checks the tailroom on every RFCOMM PDU allocation and closes this residual write. EPSS 0.4%Vulnerability | CVE Database V5 | ||
| Medium | 5.8 | EPSS 0.1% | CVE-2026-102938pypa virtualenv · CRLF Injection Medium5.8EPSS 0.1%CVE Database V5CVE-2026-102938 ### Summary `pyvenv.cfg` is a line-based format with no escape syntax. `PyEnvCfg.write()` wrote values verbatim, while `PyEnvCfg._read_values()` parses the file with `str.splitlines()`. A value containing a line boundary therefore became additional configuration lines, and because reading is last-wins, the injected keys replaced any key written earlier in the file. ### Impact The `prompt` value is the reachable input: it is set by `--prompt`, by the `VIRTUALENV_PROMPT` environment variable, or from the config file, and `write()` emits `prompt` before `home`. A crafted prompt can therefore set `home` in the generated `pyvenv.cfg`: ```console $ virtualenv --prompt $'x"\nhome = /attacker/path\nprompt = "z' venv $ grep '^home' venv/pyvenv.cfg home = /attacker/path ``` `home` is what tooling reads to locate the base interpreter, so a consumer that trusts it can be pointed elsewhere. `implementation`, `version_info`, `version`, `executable`, `command` and `virtualenv` are also written before `prompt` and can be replaced the same way. This requires the prompt to come from somewhere other than the person running the command, for example a CI job templating a branch name into it, tooling deriving an environment name from user-supplied data, or an inherited `VIRTUALENV_PROMPT`. Where the operator supplies the prompt directly they already control the command line, and the effect is corruption rather than privilege gain: the value is truncated at the boundary and read back with a dangling quote. ### Details The boundary set is the one `str.splitlines()` recognizes, which is wider than `\n`: `\r`, `\v`, `\f`, the file, group and record separators, `U+0085`, `U+2028` and `U+2029` were all written through unchanged and all split the line when read back. ### Patches `PyEnvCfg.write()` now collapses those boundaries to spaces as it serializes each line, so it cannot emit a structurally invalid file regardless of what a caller places in `content`. ### Workarounds Do not pass externally influenced data as the virtualenv prompt. Strip line boundaries from any value before using it as `--prompt` or `VIRTUALENV_PROMPT`. | CVE Database V5 | ||
| High | 7.8 | EPSS 0.2% | High7.8EPSS 0.2%GCVE DatabaseCVE-2026-80961 In the Linux kernel, the following vulnerability has been resolved: dm-pcache: validate kset key_num and intra-segment bounds Two more fields decoded from the cache device go unbounded. The kset key_num drives cache_kset_crc() and the replay loop in cache_replay(), the writeback worker and the GC worker, but only the magic and a fixed-seed CRC are checked first, so a non-last kset whose key_num exceeds the PCACHE_KSET_KEYS_MAX buffer reads past its end before the CRC compare. A key's intra-segment offset and length in cache_key_decode() are taken verbatim, so a key running past its segment is replayed into the cache tree and the data CRC check and every later read hit then copy adjacent persistent memory into the caller's bio -- an out-of-bounds read that leaks to user space. Both fields are controlled by whoever supplies the cache device (CAP_SYS_ADMIN); the CRC seed is public. Add kset_onmedia_valid() to bound key_num before any kset read, and reject a key whose offset plus length, computed in 64 bits, exceeds the segment data_size. Valid metadata is unaffected. | GCVE Database | ||
| Medium | – | EPSS 0.2% | MediumEPSS 0.2%GCVE DatabaseCVE-2026-80778 In the Linux kernel, the following vulnerability has been resolved: futex/pi: Reject cross-mm private futex owners A private futex key borrows the waiter's mm without taking an mm_users reference. Nevertheless, attach_to_pi_owner() currently accepts an owner from a different address space and copies the private key into the owner's PI state. When that owner exits, exit_pi_state_list() uses the saved key to find the hash bucket and acquires a reference to the waiter's private hash. If the last user of the waiter's mm exits concurrently, futex_hash_free() frees the hash while the owner still uses its bucket and reference. Prevent this by validating in attach_to_pi_owner() that, for private futexes, the owner mm and waiter mm are the same. Perform the check with the owner's pi_lock held and after validating owner::futex::state to serialize against a concurrent PI-state exit cleanup. [ tglx: Amended comment ] | GCVE Database | ||
| Critical | 9.8 | EPSS 0.4% | Critical9.8EPSS 0.4%GCVE DatabaseCVE-2026-80519 In the Linux kernel, the following vulnerability has been resolved: ovpn: finish crypto callback cleanup before peer release Crypto completion callbacks hold both key-slot and peer references. The peer reference pins the netdev, and dropping the last peer reference can let netdev unregistration and module removal make progress. Do not release that peer reference before the callback has finished its own cleanup. If ovpn_crypto_key_slot_put runs after ovpn_peer_put, it can schedule an RCU callback backed by module text after ovpn_cleanup rcu_barrier has already run. The TX error path also freed the remaining skb after ovpn_peer_put, leaving callback cleanup outside the peer/netdev lifetime window. Release the key slot and free any remaining skb first, then drop the peer reference as the last callback action. | GCVE Database | ||
| Medium | 5.3 | EPSS 0.4% | Medium5.3EPSS 0.4%CVE Database V5CVE-2026-77780 CVE-2026-77780 is an authorization bypass vulnerability in Roskus Prospero Flow CRM versions 4.9.1 through before 5.14.2. It allows a user with transaction and accounting creation permissions to access sensitive bank account information of other companies by supplying a bank_account_id or bank_card_id in the transaction save endpoint. The sensitive data disclosed includes bank account name, bank name, and the last four digits of the card. This occurs because the application does not verify company ownership before persisting and rendering this information. | CVE Database V5 | ||
| High | 7.7 | EPSS 0.4% | CVE-2026-76216vikunja · Insecure Direct Object Reference High7.7EPSS 0.4%CVE Database V5CVE-2026-76216 ## Summary Vikunja's `web.Auth` interface (`pkg/web/web.go`, single method `GetID() int64`) is satisfied by BOTH `*user.User` and `*models.LinkSharing`. A link-share's `GetID()` returns the **raw positive** `share.ID` (`pkg/models/link_sharing.go:83-85`), which lives in the same positive autoincrement ID space as `users.id`. The safe negated form `getUserID() = share.ID * -1` (`link_sharing.go:126-128`) exists but is NOT used at three permission sinks. As a result, a link-share principal with id `N` — which should have zero authority over teams or bot users — is treated as the *user* whose `users.id == N` at three permission checks that lack the `a.(*LinkSharing)` guard their sibling methods have. This is the same principal-type-confusion class as CVE-2026-68581 (GHSA-vvcv-vpph-h844), but at three code paths that advisory/fix never touched. ## Root Cause `web.Auth` is a one-method interface (`GetID() int64`). `*LinkSharing.GetID()` returns the raw positive share id. Three permission methods compare this raw id directly and omit the link-share type guard used elsewhere in the same files: 1. **`TeamMember.CanDelete`** (`pkg/models/team_members_permissions.go:31-40`): the self-removal fast path `if u.ID == a.GetID() { return true }` (:36) executes **before** `IsAdmin`. `IsAdmin` (:48-51) is the ONLY place that rejects link shares (`if _, is := a.(*LinkSharing); is { return false }`, :50) — and it is never reached when the fast path returns true. 2. **`BotUser.isOwner`** (`pkg/models/bot_users_permissions.go:47-56`): `return u.BotOwnerID == a.GetID()` (:55), used by `CanRead`/`CanUpdate`/`CanDelete` (:36-45). Unlike `CanCreate` (:27-30) which type-asserts `a.(*user.User)`, these three paths have no principal guard. 3. **`Team.CanRead`** (`pkg/models/teams_permissions.go:68-78`): matches membership on `And("user_id = ?", a.GetID())` (:76) with no link-share guard, unlike sibling `IsAdmin` (:45-49, guard at :47). Link-share JWTs reach these routes: `SetupTokenMiddleware` validates the signature only, `GetAuthFromClaims` returns `*models.LinkSharing`, and the `/user`/`/teams` route groups add no link-share rejection. Link sharing is enabled by default (`config.go` `ServiceEnableLinkSharing.setDefault(true)`). ## Impact A link-share principal (obtainable from any public share link, or self-registered via a share on the attacker's own project) whose id `N` collides with a victim's `users.id == N` can, without being that user or any user: - **Integrity (I:H):** remove the victim from any team they belong to (`DELETE /api/v1/teams/{T}/members/{username}`) → revokes all project permissions the victim inherited through that team. - **Availability/Integrity (A:H, I:H):** enumerate the victim's bot users (`GET /api/v1/user/bots` → `bot_owner_id = a.GetID()`), then disable/rename or **permanently delete** them (`DELETE /api/v1/user/bots/{id}` → `DeleteUser`), destroying data owned solely by those bots. - **Confidentiality (C:H):** read the roster + metadata (name, description, full member list) of any team the colliding user belongs to (`GET /api/v1/teams/{T}`), plus read bot-user records via `isOwner`-gated reads. ## Attack Chain ### Sink 1 — TeamMember.CanDelete (integrity) 1. **Entry:** `POST /api/v1/shares/{hash}/auth` → link-share JWT with `id = N`. **Guard:** JWT middleware — signature only. **Bypass proof:** `GetAuthFromClaims` returns `*models.LinkSharing`; no route-group link-share rejection on the `/teams` group. 2. **Action:** `DELETE /api/v1/teams/{T}/members/{usernameOfUserN}` with the link-share bearer, targeting a team `T` (≥2 members) that user `N` belongs to. **Guard:** `CanDelete` → `GetUserByUsername(tm.Username)` returns user N, then `u.ID == a.GetID()` (`team_members_permissions.go:36`). **Bypass proof:** `a.GetID()` returns `N` (raw positive `share.ID`, `link_sharing.go:84`) == user N's id → `true`. No `a.(*LinkSharing)` check on this branch (only `IsAdmin` at :50 has it, never reached). 3. **Sink:** `Delete` (`team_members.go`) removes user N from team T (last-member check passes when team has ≥2 members). **Impact:** victim loses all project access inherited through team T. ### Sink 2 — BotUser.isOwner (bot takeover / destruction) 1. **Entry:** link-share JWT id N (as above). **Guard:** signature-only; `/user` group adds no link-share reject. 2. **Enumerate:** `GET /api/v1/user/bots` → `ReadAll` runs `Where("bot_owner_id = ?", a.GetID())` = bots owned by user N. **Bypass proof:** `a.GetID()` = N; returns victim's bot ids self-contained (removes the id-guessing barrier). 3. **Sink:** `DELETE /api/v1/user/bots/{botId}` → `CanDelete` → `isOwner` → `u.BotOwnerID == a.GetID()` (`bot_users_permissions.go:55`) → true; `Delete` calls `DeleteUser`. **Bypass proof:** no `a.(*LinkSharing)` guard here (Create-only, :28). **Impact:** disable/rename/permanently delete victim's bot automation identities. ### Sink 3 — Team.CanRead (info disclosure) 1. **Entry:** link-share JWT id N. **Guard:** signature-onl | CVE Database V5 | ||
| Medium | 6.9 | EPSS 0.1% | CVE-2026-74871jahlives openssl_encrypt · Weak Password Hashing Medium6.9EPSS 0.1%CVE Database V5CVE-2026-74871 openssl_encrypt versions before 1.4.6 contain a key derivation flaw in sequential XOR composition mode where the last stage cancels out during key generation. When configured with a single KDF and no prior hashing stage, attackers can bypass memory-hard key derivation and perform offline password cracking at SHA-256 speed instead of the configured KDF cost. | CVE Database V5 | ||
| Low | 1.8 | EPSS 0.2% | Low1.8EPSS 0.2%GCVE DatabaseCVE-2025-47278 CVE-2025-47278 is a low-severity vulnerability in Flask 3.1.0 affecting the icloudpd product versions >=1.25.1 and <1.28.1_1. The issue arises from incorrect handling of fallback key configuration for session signing, causing the last fallback key to be used instead of the current signing key. This impacts sites using key rotation via SECRET_KEY_FALLBACKS, leading to sessions being signed with stale keys, which may impede key transitions but does not compromise data integrity. | GCVE Database | ||
| Critical | 9.8 | EPSS 0.7% | CVE-2026-16770MITHALDU PDF::WebKit · Argument Injection Critical9.8EPSS 0.7%CVE Database V5CVE-2026-16770 PDF::WebKit versions through 1.2 for Perl allow argument injection into wkhtmltopdf via meta tags in the source document. For an HTML string or file source, the constructor collects every <meta name="pdf-webkit-KEY" content="VALUE"> element in the document head through _pdf_webkit_meta_tags and turns each one into a wkhtmltopdf command line option. KEY is normalized to an option name matching --[a-z0-9-]+ but is not checked against an allow list, VALUE is passed through unchanged as the argument that follows it, and a VALUE of "yes" emits the option as a bare flag. BUILD merges the meta derived options last, so they also override the module defaults and the options passed to new. Switches such as --enable-local-file-access and --cookie-jar are reachable this way. The renderer is executed with an argument list rather than a shell command, so this is argument injection and not shell injection. Any caller that renders untrusted HTML lets the document choose the renderer's options and override those set by the application, including options that read local files into the resulting PDF or write to a chosen path. A URL source is not scanned, and the scan is skipped when XML::LibXML, a recommended dependency, is not installed. | CVE Database V5 | ||
| –Not rated | – | Bleeping Computer In late 2025, a small combined heat-and-power (CHP) plant in Poland was breached by attackers using a private APN to access its operational technology (OT) network. The attackers exploited a misconfiguration that allowed devices within the private APN network to communicate freely, enabling lateral movement. They compromised a FortiGate VPN/firewall and a Teltonika cellular router to tunnel into the private APN, then accessed a WAGO PFC200 PLC with default credentials. The attackers subsequently disabled key systems including a steam turbine and water treatment system by switching PLCs to STOP mode and activating password protection. The outage was short-lived and did not impact the population. This incident is notable as the first known real-world cyberattack leveraging lateral movement through a private APN to reach OT networks. The Polish CERT recommends treating private APNs as untrusted external networks and implementing client isolation and traffic allowlists. | Bleeping Computer | |||
| High | 7.8 | EPSS <0.1% | High7.8EPSS <0.1%GCVE DatabaseCVE-2026-68148 In the Linux kernel, the following vulnerability has been resolved: fscrypt: Add missing superblock check in find_or_insert_direct_key() The legacy 'fscrypt_direct_keys' table caches master keys that are used by v1 encryption policies that have FSCRYPT_POLICY_FLAG_DIRECT_KEY. It's just a global table for all filesystems (since the keys can be provided by the legacy process-subscribed keyrings mechanism, which makes it difficult to reuse super_block::s_master_keys). The entries in it ('struct fscrypt_direct_key') do contain a super_block pointer, though, for passing to fscrypt_destroy_inline_crypt_key() when the last inode that references the key is evicted. However, when finding the fscrypt_direct_key for an inode, we weren't actually comparing the super_block pointer. As a result, inodes with different super_blocks could point to the same fscrypt_direct_key. That could extend the lifetime of a fscrypt_direct_key beyond the super_block it points to, causing a use-after-free later. Fix this by creating distinct fscrypt_direct_key structs for distinct super_block structs. Note that this problem doesn't exist in the v2 policy equivalent ("per-mode keys"), since the data structures there are per super_block. | GCVE Database | ||
| High | 7.8 | EPSS 0.2% | High7.8EPSS 0.2%GCVE DatabaseCVE-2026-64363 In the Linux kernel, the following vulnerability has been resolved: HID: appleir: fix UAF on pending key_up_timer in remove() appleir_remove() runs hid_hw_stop() before timer_delete_sync(). hid_hw_stop() synchronously unregisters the HID input device via hid_disconnect() -> hidinput_disconnect() -> input_unregister_device(), which drops the last reference and frees the underlying input_dev when no userspace handle holds it open. key_up_tick() reads appleir->input_dev and calls input_report_key() / input_sync() on it. The timer is armed from appleir_raw_event() with a HZ/8 (~125 ms) timeout on every keydown and key-repeat report. If a key was pressed shortly before the device is disconnected, the timer can fire after hid_hw_stop() has freed input_dev but before the teardown drains it. A simple reorder is not sufficient. Putting the timer drain first still leaves a window where a USB URB completion (raw_event) running during hid_hw_stop() can call mod_timer() and re-arm the timer, which then fires after hidinput_disconnect() has freed input_dev. The same URB-completion window also lets raw_event() reach key_up(), key_down() and battery_flat() directly, all of which dereference appleir->input_dev. Introduce a 'removing' flag on struct appleir, gated by the existing spinlock. appleir_remove() sets the flag under the lock and then shuts down the timer with timer_shutdown_sync(), which both drains any in-flight callback and permanently disables further mod_timer() calls. appleir_raw_event() and key_up_tick() bail out early if the flag is set, so no path can arm or run the timer, or dereference appleir->input_dev, after remove() has started tearing down. The keyrepeat and flatbattery branches of appleir_raw_event() previously called into the input layer without holding the spinlock; take it now so the flag check is well-defined. This incidentally closes a pre-existing read-side race on appleir->current_key in the keyrepeat branch. This bug is structurally a sibling of commit 4db2af929279 ("HID: appletb-kbd: fix UAF in inactivity-timer cleanup path") and has been present since the driver was introduced. | GCVE Database | ||
| Medium | – | MediumCisco Talos Welcome to this week’s edition of the Threat Source newsletter. Lately I've found myself thinking a lot about the Australian TV series Mr. Inbetween (IMDb 8.7/10) — not because I'm a hitman for hire, but because I literally feel in-between. Specifically, in-between what I'd call the "pre-Mythos" and “post-Mythos” eras. We've crossed a capability threshold, and it's not just one model family driving that — Codex 5.3 and GPT-5.5 deliver comparable or better performance , and Tulongfeng or GLM-5.2 (an incredibly powerful open-weight model, MIT-licensed) show the frontier isn't limited to closed models anymore. On the other side of that line, real-world impact hasn't caught up yet and we're living in an artificial buffer zone. For me, defining the “pre-” and “post-” status comes down to the gap between "vulnerability discovery" and "vulnerability publication." Last week’s Patch Tuesday gave a signal of change, as Joe pointed out , so maybe the buffer zone has come to an end. Let's dive into the 2026 Q2 stats. As in past years, we're seeing a steeper curve than the year before — a solid 49% YoY growth, though still not the hockey-stick moment I keep waiting for. By the end of June we were tracking close to 200 CVEs per day. Using the keyword methodology described here , I found 452 AI-related CVEs this calendar year. If "openclaw" is added to the keyword list, that number jumps by another 536 — a reminder that these counts are sensitive to keyword drift. Given how much the keyword list keeps changing, I'm reconsidering whether to keep publishing this particular metric going forward. KEVs, by contrast, "only" grew 13% — a small April spike aside, it's fairly flat relative to total CVE growth. Networking-gear-related CVEs continued their climb, now accounting for 24% of KEV-related vulnerabilities (up from 20% in Q1) — consistent with the trend I flagged last quarters. As in previous quarters, CVEs from 2024 or earlier still make up about 24% of everything we're tracking. More strikingly, even though the standard enterprise patch cycle is described to run 30–90 days, 181 days into 2026, 46% of today's actively-exploited (KEV) CVEs still trace back to 2025 or earlier. Old vulnerabilities don't retire, new ones keep arriving, and machine-speed vulnerability discovery is going to keep outpacing human-speed patching. Which brings me back — once again — to EPSS as a tool for prioritizing patching against this dataset. If you patched purely by CVSS 9+, you'd be urgently chasing ~3,700 CVEs — but 95% of those sit below 5% EPSS, meaning the real-world odds of exploitation are tiny. Of the 32 CVSS 9+ CVEs with EPSS ≥ 50%, 25 are already on CISA's KEV list. The remaining seven outliers are still high-probability by EPSS but haven't made KEV yet — worth watching. Ray Shoesmith (Mr. Inbetween) once told his therapist, "You know, if I hit somebody, I generally got a pretty good reason." Same principle applies to patching. Don't swing at everything — swing at what you have good reason to believe is coming for you. The one big thing Cisco Talos has discovered "msaRAT," a new Rust-based remote access trojan (RAT) deployed by the Chaos ransomware group. Built on the Tokio asynchronous runtime, it establishes a covert command-and-control (C2) channel by hijacking Chrome or Edge browsers via the Chrome DevTools Protocol (CDP). The infection starts with a deceptive MSI file masquerading as a Windows update that loads the payload directly into memory, paving the way for ransomware deployment. Why do I care? This RAT is a master of evasion, living off the browser to build its C2 infrastructure without ever directly touching the network. By routing traffic through legitimate browser processes and trusted services, msaRAT easily bypasses traditional network-based detections. Additionally, its use of the Tokio runtime enables highly efficient, parallel execution of malicious tasks, accelerating the attacker's ability to establish persistence and deploy doubl… | Cisco Talos | |||
| Critical | 10.0 | EPSS 0.6% | CVE-2026-44182jupyter-server enterprise_gateway · Injection Critical10.0EPSS 0.6%CVE Database V5CVE-2026-44182 Jupyter Enterprise Gateway launches remote Jupyter Notebook kernels across distributed clusters like Apache Spark, Kubernetes, and Docker Swarm. In versions prior to 3.3.0, the server interpolates untrusted environment variables (e.g., KERNEL_XXX) into Kubernetes manifests without YAML-aware escaping, enabling YAML injection attacks. Attackers can inject new fields, overwrite critical fields (e.g., duplicate securityContext keys, where the last one prevails), and inject document boundaries (--- for new documents, ... for end-of-document) to generate multiple resources, potentially creating arbitrary types, such as privileged pods. The Jinja2 template for the Kubernetes manifest contains several kernel_xxx variables, such as kernel_working_dir that are used when rendering the manifest and are all vectors for YAML injection. This issue has been fixed in version 3.3.0. | CVE Database V5 | ||
| –Not rated | – | Check Point Research For years, the cyber security industry tracked AI as a force multiplier: something that made existing attack techniques faster, cheaper, and more accessible. That framing was accurate. But the Annual AI Security Report 2026 from Check Point Research documents a transition that goes further. AI has crossed from assistant to operator. Where it once helped attackers prepare, it now runs the operation. Key observed findings AI has crossed from development aid to live attack operator. It now does the hands-on work inside live intrusions, from China-nexus espionage campaigns to a criminal breach of multiple Mexican government agencies and has spread from nation states to ordinary cyber criminals. AI now builds deployment-ready malware and attack suites. Its involvement is often invisible in the finished artifact: one developer used an AI environment to produce VoidLink, an 88,000-line command-and-control offensive framework, in under a week. Attackers prefer commercial models, and now abuse them by exploiting the agentic architecture, not just single prompts. Most actors favor jailbroken mainstream models over self-hosted ones, and the durable bypass is now a planted configuration file an agent loads and trusts across sessions. An AI-enabled criminal tooling market has matured. Phishing-as-a-service kits now embed a language model with the jailbreak built in, and conversational AI voice-agent services run vishing and one-time-passcode theft at scale. Virtual Identity is no longer a reliable trust anchor. Voice, face, documents, and live video are now cheap to forge convincingly and are widely used in attacks taking multi-channel social engineering to a new level of integration. AI itself is an expanding attack surface. Models cannot always separate data from instructions and content they process might influence the model’s behavior; the surrounding stack adds ordinary software vulnerabilities and supply-chain risk, all in a rapidly evolving ecosystem where security practices not always mature. Indirect prompt injection is on the rise. Detections of longer malicious payloads increased sharply, rising roughly fivefold between March and May 2026 and approaching 1% of observed prompts in May. Longer payloads are more typical of content-borne and agentic attack paths, this pattern suggests that indirect prompt injection is becoming more operationally relevant. Enterprise data leakage through GenAI is persistent and growing risk . High-risk prompts doubled from 2% to 4% during the last year, while organizations used an average of 10 AI applications each month, many without official approval. Data exposure risks are not evenly distributed across the verticals . Sector-level analysis reveals that AI-related data exposure risks are not evenly distributed across the verticals, and correlate both with AI usage patterns and security maturity. Business Services recorded the highest rate of high-risk GenAI prompts at 5.91%, meaning nearly one in every 17 AI interactions carried a significant risk of sensitive data exposure. To read the full findings, access the AI Security Report 2026 from Check Point Research here. The post AI Security Report 2026 appeared first on Check Point Research . | Check Point Research | |||
| Malicious | – | MaliciousGCVE Database The npm package 'tipsen-last-pls' contains malicious code that can fully compromise any computer on which it is installed or running. Removal of the package does not guarantee elimination of all malicious software, as the attacker may have gained full control of the system. Immediate rotation of all secrets and keys stored on the affected computer from a different, secure device is strongly advised. | GCVE Database | |||
| Malicious | – | MaliciousGCVE Database The npm package 'tipsen-last' contains malicious code that can fully compromise any computer on which it is installed or running. The presence of this package indicates a severe security breach, and all secrets and keys stored on the affected system should be rotated immediately from a different, uncompromised machine. Removing the package alone may not eliminate the malicious software introduced, as full control may have been granted to an attacker. | GCVE Database | |||
| –Not rated | – | Krebs on Security Two men pleaded guilty in the United Kingdom this week to criminal charges stemming from an August 2024 cyberattack that crippled Transport for London , the entity responsible for the public transport network in the Greater London area. The duo were key members of a prolific cybercrime group known as Scattered Spider , and their guilty pleas came on the first day of what was expected to be a six-week trial. Owen Flowers (left) 18, and Thalha Jubair, 20. Image: UK National Crime Agency (NCA). Thalha Jubair , 20, of East London and 18-year-old Owen Flowers of Walsall admitted conspiring to commit unauthorized acts against Transport for London computer systems and causing risk of serious damage to human welfare. According to a report from the BBC, Flowers alone admitted to being part of a conspiracy to hack into U.S. based healthcare providers SSM Health Care Corporation and Sutter Health in September 2024. Jubair is also wanted by U.S. law enforcement agencies. In September 2025, prosecutors in New Jersey unsealed an indictment alleging Jubair and other Scattered Spider members committed computer fraud, wire fraud, and money laundering in relation to 120 computer network intrusions involving 47 U.S. entities between May 2022 and September 2025, and that the group’s victims paid at least $115 million in ransom payments. In July 2025, KrebsOnSecurity reported that Flowers and Jubair were arrested in the United Kingdom in connection with Scattered Spider ransom attacks against the retailers Marks & Spencer and Harrods , and the British food retailer Co-op Group . Multiple sources familiar with those investigations said Flowers was the Scattered Spider member who anonymously gave interviews to the media in the days after the group’s September 2023 ransomware attacks disrupted operations at Las Vegas casinos operated by MGM Resorts and Caesars Entertainment . According to prosecutors, Jubair co-ran a bustling Telegram channel called Star Chat , the home of a SIM-swapping group that used voice- and SMS-based phishing attacks to steal credentials from employees at the major wireless providers in the U.S. and U.K. The group would then use that access to sell a service that could redirect a target’s phone number to a device the attackers controlled and intercept the victim’s calls and text messages (including one-time codes for multi-factor authentication). A receipt from Star Fraud Chat’s SIM-swapping service targeting a T-Mobile customer after the group gained access to internal T-Mobile employee tools. “Rocket Ace” was one of Jubair’s hacker handles, according to U.S. prosecutors. New Jersey prosecutors also allege Jubair also was involved in a mass SMS phishing campaign during the summer of 2022 that stole single sign-on credentials from employees at hundreds of companies. That weeks-long SMS phishing campaign led to intrusions and data thefts at more than 130 organizations, including LastPass , DoorDash , Mailchimp , Plex and Signal . KrebsOnSecurity reported last year that one of Jubair’s alter egos at age 15 was “ Everlynn ,” a hacker who sold fraudulent “emergency data requests” that used compromised police and government email addresses to demand subscriber data (e.g. username, IP/email address) from major tech companies, claiming the requests concerned urgent matters of life and death and could not wait for a court order. In April 2026, 24-year-old British national and Scattered Spider member Tyler “Tylerb” Buchanan pleaded guilty to wire fraud conspiracy and aggravated identity theft for participating in the group’s SMS phishing spree in the summer of 2022. The government said Buchanan, Jubair and others used the credentials harvested in that phishing campaign to steal at least $8 million in cryptocurrency from victims throughout the United States. Buchanan is currently scheduled to be sentenced on October 2. In August 2025, 20-year-old Scattered Spider member from Florida named Noah Michael Urban was sentenced to 10 years in… | Krebs on Security | |||
| Critical | 9.1 | Critical9.1GCVE Database ## Summary `JWSVerifier::getAlgorithm()` in `src/Library/Signature/JWSVerifier.php` (line 144) merges protected and unprotected headers using PHP's spread operator: ```php $completeHeader = [...$signature->getProtectedHeader(), ...$signature->getHeader()]; ``` In PHP, when spreading arrays with duplicate string keys, the **last array's values take precedence**. Since the unprotected header (`getHeader()`) is spread second, an attacker can override the integrity-protected `alg` parameter by placing a different value in the unprotected header. This creates a Time-of-Check/Time-of-Use (TOCTOU) vulnerability: 1. `HeaderCheckerManager` validates `alg` from the **protected** header 2. `JWSVerifier` uses `alg` from the **unprotected** header for actual verification The same issue exists in `JWEDecrypter.php` (lines 120-124) where `array_merge()` exhibits the same last-wins behavior for `alg` and `enc`. ## Affected Code **JWSVerifier.php line 144** — Spread operator merge order allows unprotected header to override `alg`: ```php $completeHeader = [...$signature->getProtectedHeader(), ...$signature->getHeader()]; ``` **JWEDecrypter.php lines 120-124** — `array_merge()` with same last-wins behavior: ```php $completeHeader = array_merge( $jwe->getSharedProtectedHeader(), $jwe->getSharedHeader(), $recipient->getHeader() ); ``` ## Attack Vectors ### Vector A — Mixed key sets (HIGH probability) If the application uses a JWKSet containing keys of different types (common in multi-tenant or federation scenarios), the JWSVerifier iterates all keys (line 86). An attacker can force a different algorithm that matches a different key in the set. ### Vector B — alg ONLY in unprotected header (HIGH probability) If `alg` is placed EXCLUSIVELY in the unprotected header (not in the protected header at all), `HeaderCheckerManager::checkDuplicatedHeaderParameters()` does NOT trigger. The JSON Flattened/General serializers allow tokens with no protected header or a protected header without `alg`. RFC 7515 Section 4.1.1 states `alg` MUST be integrity-protected, but the library does not enforce this. ### Vector C — Direct JWSVerifier usage (HIGH probability) `JWSLoader` takes `?HeaderCheckerManager` (nullable). If developers use `JWSVerifier` directly or create `JWSLoader` without a `HeaderCheckerManager`, the duplicate header check never runs. ## Contrast with JWSBuilder (safe) `JWSBuilder::findSignatureAlgorithm()` (line 196) uses `[...$header, ...$protectedHeader]` where protected wins. It also has `checkDuplicatedHeaderParameters()` (line 218). The JWSVerifier has **neither** safeguard. ## Proof of Concept ```php <?php // Demonstrate algorithm override via unprotected header $protected = ["alg" => "RS256", "typ" => "JWT"]; $unprotected = ["alg" => "HS256"]; $merged = [...$protected, ...$unprotected]; // $merged["alg"] === "HS256" — unprotected wins! // JSON Flattened JWS with algorithm override: $maliciousJws = json_encode([ 'payload' => base64url_encode($payload), 'protected' => base64url_encode('{"alg":"RS256"}'), 'header' => ['alg' => 'HS256'], // OVERRIDE 'signature' => base64url_encode($sig), ]); // HeaderCheckerManager validates RS256 from protected header -> PASS // JWSVerifier uses HS256 from unprotected header -> attacker's algorithm choice ``` A full working PoC demonstrating HS512-to-HS256 downgrade with mixed keysets is available upon request. ## Suggested Fix In `JWSVerifier::getAlgorithm()`, read `alg` exclusively from the protected header: ```php private function getAlgorithm(Signature $signature): Algorithm { $protectedHeader = $signature->getProtectedHeader(); if (! isset($protectedHeader['alg'])) { throw new InvalidArgumentException('The "alg" parameter must be in the protected header.'); } return $this->signatureAlgorithmManager->get($protectedHeader['alg']); } ``` For `JWEDecrypter`, reverse the merge order so protected header wins, or extract `alg`/`enc` exclusively from the protected header. ## Résolution Un correctif a été préparé sur une branche dédiée basée sur `3.4.x`, avec des tests anti-régression dédiés (fork privé temporaire de cette advisory, PR #1). **JWS algorithm confusion** — `JWSVerifier` lit le paramètre `alg` exclusivement dans le header protégé en intégrité (RFC 7515 §4.1.1) ; un `alg` placé dans le header non protégé ne peut plus surcharger l'algorithme signé. **Validation :** `php -l` OK, PHPUnit vert, aucune nouvelle erreur PHPStan introduite (différentiel nul vs `3.4.x`), aucun commentaire ajouté dans le code source. Après merge, cascade prévue `3.4.x → 4.0.x → 4.1.x`. | GCVE Database |
Page 1 of 2