The Briefing

What happened, why it matters, and what to do about it. Analysis, not rewritten press releases.

Showing 61 to 96 of 654 briefings

Thursday 22 October

A dark editorial scene for the briefing, with its headline and key figure drawn over it.

Threat Intel22 Oct 2026

Kaspersky says Lazarus malware uses ML-KEM-1024 key exchange in Operation FlashHole against South Korea

What the Securelist report on Operation FlashHole says about ML-KEM-1024 in LPEClient, the infection chain, victims and indicators, checked against the report, the Kaspersky release and NIST FIPS 203.

A dark panel with a notice that this briefing has been withdrawn for rewriting.

Threat Intel22 Oct 2026

SafePay advisory AA26-293A briefing withdrawn: most of its figures and technical detail are not in the advisory

Why this SafePay briefing was withdrawn, and what the CISA and FBI advisory AA26-293A actually says.

AI Security22 Oct 2026

Claude Duet briefing withdrawn: it described a different product from the one Anthropic announced

Why this Claude Duet briefing was withdrawn, and what Anthropic says Claude Duet is.

Market and Supply Chain22 Oct 2026

LiteLLM 1.107.0 compromise: this briefing is withdrawn; its figures, indicators and commands were not supported

Why this LiteLLM briefing was withdrawn, what LiteLLM's own posts say about the 21 October compromise, and where the corrected briefing is.

Wednesday 21 October

AI Security21 Oct 2026

NVIDIA confidential computing briefing withdrawn: its headline overhead figure and library have no vendor source

Why this NVIDIA confidential computing briefing was withdrawn, and what NVIDIA and Google Cloud actually announced on 21 October.

AI Security21 Oct 2026

Cohere Rerank 5 briefing withdrawn: the security boundary it described is not in Cohere's announcement

Why this Cohere Rerank 5 briefing was withdrawn, and what Cohere's own pages say about the 21 October release.

Market and Supply Chain21 Oct 2026

Google Cloud and NVIDIA say A5X Vera Rubin instances reach GA next month, with up to 960,000 GPUs planned

What Google Cloud and NVIDIA published on 21 October 2026 about A5X instances on Vera Rubin NVL72, what NVIDIA's datasheet lists for memory, NVLink, rack power and cooling, and what neither company states.

AI Security21 Oct 2026

OpenAI details Patch the Planet and a five-stage response framework after the PHP2Shell flaws

What OpenAI says about its Patch the Planet programme, built with Trail of Bits, and about its role in the PHP2Shell response; its proposed five-stage framework; what the post does not say; and how the PHP Foundation and Cloudflare describe the leak.

A cinematic data center at night with rows of metallic server racks and illuminated green and cyan fiber optic status indicators on the right. The left half is darkened with overlay headline text describing the emergency PHP remote code execution fixes and a stat highlighting 22.6 million exposed hosts.

Threat Intel21 Oct 2026

PHP issues emergency fixes for two critical RCE flaws: CISA and UK NCSC issue urgent alerts

What PHP's advisory, the NCSC and CISA say about the PHP2Shell flaws: the four fixed releases, the stopgaps PHP prints, who is affected, and a corrected update checklist.

Editorial illustration of an unbranded server maintenance desk with empty update rows on two screens. The headline states that PHP has two critical security fixes, each rated 9.2 out of 10.

Threat Intel21 Oct 2026

PHP patches two critical RCE flaws: mod_php highest risk, FPM needs attention too

Two PHP 9.2 flaws can lead to unauthenticated code execution. The official advisories distinguish mod_php and FPM exploit evidence, fixed versions and interim controls.

Editorial illustration of an unbranded enterprise firewall and backup storage rack at night. The headline says SafePay, 600 victims and the two entry routes.

Threat Intel21 Oct 2026

SafePay: agencies count at least 600 victims, but the access route varies

The joint SafePay warning identifies at least 600 victims and two distinct access families. Here is how to test each exposure without treating the campaign as one fixed attack chain.

Editorial image of an unbranded laptop with a generic blank order receipt and a phone beside it. Text says a genuine sender can carry a planted phone lure, noting Talos's roughly 1,400-message sample.

Threat Intel21 Oct 2026

The order email passed authentication. The planted helpline was the attack

In Talos's samples, the sender passed checks while a fake support number appeared in shipping or billing fields. We trace that boundary, the 1,400-message denominator and two related notification routes.

Page 7

Wednesday 21 October

AI-assisted editorial illustration of server modules being moved inside a data centre. Text says 16 managed AI model IDs reach their retirement date and notes that no outage was observed here.

Market and Supply Chain21 Oct 2026

Google Cloud's 16 managed model IDs reach retirement date. Find the one in production

The complete 16-ID inventory, what Google's retirement wording means, where self-deployment differs from MaaS, and the tests to run before routing production traffic elsewhere.

Editorial illustration of a studio microphone and audio waveform. Text reads Voxtral 3, independent test records 2.1% word error, live-agent limits still matter.

AI Security21 Oct 2026

Voxtral 3 scores 2.1% on recorded speech. The voice agent is a separate test

Voxtral Small Transcribe 3 records 2.1% error in Artificial Analysis testing. We check the six Mistral prices, live-session limits and a conflict in TTS language documentation.

Editorial illustration of an unbranded data-centre desk with a blank table on acrylic, connected server modules and a wordless forecast trace. Headline explains NVIDIA Kumo and Agent Knowledge Foundation.

Market and Supply Chain21 Oct 2026

NVIDIA puts Kumo predictions into its agent stack. A score is not permission to act

The 21 October launch joins structured-data prediction to agent tooling. We separate the four Kumo tasks, earlier model dates, licence differences and the control needed between a prediction and an action.

Editorial illustration of network servers with a blue route diverted through a red intermediary node, representing reported SSH interception at Tor exit relays. Headline identifies 333 measured exits and 12 interception servers.

Threat Intel21 Oct 2026

333 Tor exits redirected SSH: the host-key warning was the decisive boundary

The 333 measured SSH redirects, the 377 policy matches, the Ebury case on one operator's servers and the exact point at which a host-key warning matters.

Editorial illustration of an unbranded WordPress-style plugin dashboard on a laptop, with a blank checklist and security key on the desk. Headline says UpdraftCentral keys reached customer sites; affected-site count unknown.

Breaches and Incidents21 Oct 2026

UpdraftCentral: revoked keys do not remove backdoors already planted on sites

A validly signed remote command crossed the trust boundary into connected WordPress sites. Key revocation and removal of earlier site changes are separate tasks.

Editorial illustration of an unbranded database rack and a monitor with blank search result shapes. Text on the left says Trusted Answer Search, eight security fixes, and two 9.9 scores requiring low privileges.

AI Security21 Oct 2026

Oracle patches eight Trusted Answer Search flaws; two score 9.9 but require low privileges

The eight Oracle Trusted Answer Search fixes, who can reach each flaw, why the two 9.9 scores do not mean unauthenticated access, and what version 26.1 operators should verify before patching.

Tuesday 20 October

Editorial illustration of documents transformed into typed fields and a bounded probability curve.

Market and Supply Chain20 Oct 2026

Cloudflare's Clairic returns typed values in one pass. Its own benchmarks are the first test, not the verdict

Clairic returns typed values and bounded estimates. Cloudflare's latency and cost results are vendor tests, with arithmetic and forecasting limits.

Editorial illustration of an AI agent moving through scoped permission gates in an enterprise network.

AI Security20 Oct 2026

Microsoft wants each AI agent to have its own identity and narrower authority. The delivery status matters

Microsoft describes distinct agent identities, scoped permissions and risk-driven containment. Availability and enforcement need tenant-level checks.

Editorial illustration of punch cards and an old terminal leading to a database and a partly completed test grid.

Market and Supply Chain20 Oct 2026

OpenAI's COBOL result is striking. Its test asks a narrower question than a production migration

OpenAI reports large gains in its own low-resource-language evaluations. The separate build and migration tests have different denominators.

Editorial illustration of a remote URL crossing a boundary towards a laptop browser.

AI Security20 Oct 2026

OpenCode's MCP login crossed into PowerShell: why the HTTPS check was not the fix

OpenCode patched a Windows and WSL MCP OAuth command-injection path. The HTTPS guard alone did not stop the demonstrated payload; the browser-opener update did.

Editorial illustration of exposed agent-control dashboards with branching paths towards a server; it does not depict a specific victim.

AI Security20 Oct 2026

Twenty-plus exposed Cairn panels show AI-directed intrusion work. A project is not a victim

NorForge exposed AI attack-control panels and validated some compromises. Its panel and project counts must not be mistaken for victim counts.

Editorial illustration of a cracked intermediate certificate in a chain, with a protective barrier before a browser window.

Threat Intel20 Oct 2026

Eight eMudhra CA keys were recoverable from tiny numbers. Firefox blocked them; 30,000 certificates may need migration

Mozilla found eight recoverable eMudhra intermediate keys. Firefox blocked them; root distrust now puts about 30,000 certificates into migration, according to eMudhra.

Page 8

Tuesday 20 October

A dim data hall at night: a row of tall black server rack cabinets with perforated mesh doors, receding to the right. One cabinet in the middle is dark, its indicator lights off, between cabinets lit with rows of small cyan and green lights. A small amber lamp glows on the end wall. The left half is darkened, with the headline and the stat 38 PRODUCTS drawn over it.

Breaches and Incidents20 Oct 2026

Google says a power event hit one London zone: 38 products listed, UK banks and courts report faults

What Google's status entries say, minute by minute, what its documentation and SLAs promise when one zone fails, who said what in the UK, and what the PRA, FCA, ICO and Treasury regimes require. A developing incident, time stamped.

Editorial illustration for the briefing: Bullhorn says service is restored 62 to 63 hours after its cyber incident notice; exfiltration 'not yet determined'

Breaches and Incidents20 Oct 2026

Bullhorn says service is restored 62 to 63 hours after its cyber incident notice; exfiltration 'not yet determined'

What Bullhorn's status page, FAQ and known limitations page state and do not state about its cyber incident, the credential changes, and what UK controllers and processors must do under the ICO and NCSC texts. Time stamped 20 October 2026.

A dark desk. On the right, a thick hardback book lies open with blank pages and seven blank coloured index tabs along its left edge, a black pen resting on the page. Behind it an open laptop shows a screen with seven empty outlined boxes. The left half is empty dark navy, with the headline and a stat drawn over it: 33 strands.

AI Security20 Oct 2026

NCSC's AI research chapter, 20 October 2026: 7 problems, 33 strands, and no UK code or guideline named

What the NCSC's new AI chapter asks, counted from the pages read on 20 October 2026: seven problems and 33 strands, which have a published UK answer, which words it leaves undefined, what it does not state, the UK routes to engage, and a ranked checklist.

A dark desk in low blue light. A silver laptop on the right shows a plain list of empty rounded rows, the top few tinted amber, with no writing. In front of it lies an unmarked steel padlock with its shackle open, and beside it a steel stopwatch with a plain white dial and one blue hand. The left half is empty dark navy, with the headline and a stat drawn over it: 94 minutes.

Market and Supply Chain20 Oct 2026

OVHcloud CLI: five releases swapped for up to 94 minutes on 19 October 2026, and the checksum list was swapped too

What the public issue, GitHub's release records, the Homebrew tap and OVHcloud's statements show: which files were swapped, which install routes were exposed, what immutable releases do and do not lock, what UK guidance says, and a ranked checklist. Defender level only.

A dark IT desk at night. A matte grey laptop shows a generic backup utility made only of shapes: a sidebar of empty pills, a progress bar about two thirds full, two blank buttons and three list rows. A small pop-up card outlined in amber, with a blank banner and one empty button, overlaps the lower right corner of the window. The left half is dark, with the headline and the stat 10 + 6 + 5 MIN drawn over it.

Market and Supply Chain20 Oct 2026

AOMEI Backupper's checksum was rewritten with the file in three confirmed 10, 6 and 5 minute windows, 28 May to 20 July

What Google, ManageEngine and AOMEI say about the Backupper tampering, the intervals computed from the dates, the labels that did the work of controls, what UK guidance covers and where it stops, and a ranked checklist. Defender level only.

A dark engineering workstation at night. One monitor shows an installer-style window with blank list rows and a half-filled progress bar beside a folder window of identical blank file tiles, one tile outlined in amber and tilted. A desk lamp glows amber at the right edge. The headline and figure are added by the site.

Threat Intel20 Oct 2026

Siemens Software Center CVE-2026-110936: fix added 17 October, exploitation noted 19 October

What Siemens's advisory, blog and CVE records state about CVE-2026-110936, the four days before exploitation was noted, two scores, what Romulus is, what UK guidance reads, and a ranked defender checklist. Defender level only.

Monday 19 October

A black GPU server pulled halfway out of a dark equipment rack on steel slide rails, its lid off, showing a row of eight finned graphite heatsinks and two empty recessed plates on the front panel. The left half of the image is empty dark navy, with the headline and a stat drawn over it: 510 GB, on disk, one released format.

AI Security19 Oct 2026

DeepSeek V4.1 Flash 1017 is 510 GB under MIT: 16B active, 763B stored, as read on 19 October 2026

What 510.3 GB of MIT-licensed weights need to run: memory by format against NVIDIA hardware, a 890-byte-per-token cache that only DeepSeek's own stack is stated to reach, independent numbers, where hosted prompts go, UK controls and a ranked checklist.

A dark desk at night with a silver laptop open on a code editor. In the file tree at the left of the screen are rows of blank rounded bars with small folder and file icons, one file outlined in amber. The editor pane shows blank bars like lines of code. The left of the picture is empty and dark and carries the headline: A 'font' that is a script, planted in 94 repositories, with 66 OWNERS and 15 repositories gone, 79 still reachable.

Market and Supply Chain19 Oct 2026

GhostAction stage two, 19 October 2026: 94 repositories, 66 owners, a "font" that is a script

What the second-stage count of 94 repositories and 66 owners measures, what it does not, who says GitHub blocked the first stage, what VS Code and the NCSC document, and an ordered checklist. Defender level only.

A dark room with a black server rack holding two brushed steel rack servers with blank recessed plates, and in front of it an open laptop on a black flight case. The laptop screen shows a list of five blank pale rows, the top three marked with amber dots and the bottom two with blue dots. The left half is empty dark blue, carrying the headline: five NGINX flaws, fixed 14 Oct, nginx.org rates all five medium, and the stat 3 of 5.

Threat Intel19 Oct 2026

Five NGINX flaws fixed 14 October: nginx.org says all medium, F5's CVSS 4.0 rates three critical

The labels on five NGINX flaws (nginx.org medium, F5 CVSS 4.0 critical for three), what each flaw needs, which press claims the records support, what the distribution trackers showed on 19 October, and the UK 5 day and 14 day clocks.

A dark home-office desk at night. A matte grey laptop shows a plugin list made only of shapes: five slate blue rows with empty pills and toggle switches, and one row outlined in amber with a blank amber button. A small round desk clock with tick marks and no numerals stands beside a plain mug. The left half is dark, with the headline and the stat 11 H 10 MIN drawn over it.

Market and Supply Chain19 Oct 2026

Rank Math says PRO downloads in an 11 hour 10 minute window carried malware; no count given as of 19 October

What Rank Math's notice, the CVE record and Patchstack's entry say about the PRO plugin incident, the intervals computed from them, what the notice leaves out, what a customer cannot check, the UK guidance and a ranked checklist. Defender level only.

A small charcoal four-bay network storage box with blank drive-bay doors and one tiny blue standby light stands on a worn oak shelf in a dark room, a coil of plain grey network cable beside it. The left half is empty dark blue, with the headline and a stat drawn over it: fixed 4 August, exploitation seen 17 October, 74 days, as of 19 October.

Threat Intel19 Oct 2026

Plex2Shell: a honeypot saw exploitation 1 day after the write-up and 74 days after Plex's fix

What Plex's release notes, the CVE records, CISA and the two researchers' pages establish about Plex2Shell as of 19 October: the dates to scale, four scores that miss the chain, the build that is actually fixed, what UK guidance requires, and a ranked checklist. Defender level only.

A dark control room wall carrying a grid of about twelve columns of blank, softly backlit frosted tiles under empty rounded heading pills, with one tile outlined in indigo and two neighbours dimmer. The grid sits on the right; the left is empty dark blue carrying the headline and the stat 6 of 9 drawn over it.

Threat Intel19 Oct 2026

MITRE ATT&CK v20 adds 9 Enterprise techniques; in v20 neither prompt injection one has a detection strategy

A computed diff of MITRE's version 19.2 and 20.0 STIX data: nine new Enterprise entries, six detection strategies, SecurityWeek's claims checked, the water and Mexico campaigns against their cited sources, how NCSC pages use ATT&CK, and a ranked checklist.

Show 12 more