Security

Vulnerabilities, attacks, defenses, and the security consequences of AI.

Coverage

Vulnerabilities
Cyberattacks
Prompt injection
AI defenses

Security risk / 5 hours ago / 3 min read

Meta Failed to Detect 332 AI Child-Abuse Ads in 2026, TTP Says

The September 8 investigation says advertisements promoting AI face-swapping and video apps remained on Meta platforms, sometimes using photos of real children.

Launch / 12 hours ago / 2 min read

Google Moves Chrome to Two-Week Releases to Speed Security Fixes

Chrome 153 begins the faster rhythm across desktop, iOS, and Android. Google says the change can shorten the time between known vulnerabilities and delivered fixes while accelerating feature work.

Research / Yesterday / 3 min read

Google DeepMind’s 100-Agent Math Swarm Spread Fake Proofs in 27 Minutes

The experiment’s most consequential result was not that agents found a grading flaw. It was that transparent collaboration helped both the exploit and the resistance spread—while the agents opposing it had no way to reverse the damage.

Desk statusFast moving

Published coverage45 stories

Watching now

Prompt injection / Data exposure / Model abuse / Cyber defense / Agent permissions / Supply-chain risk

Available section sponsorshipSponsor this placementReach readers while they compare the companies, products, tools, and topics shaping AI.View website sponsorships

Why it matters

Why security matters.

AI security, vulnerabilities, cyberattacks, prompt injection, data exposure, and defensive systems.

AI creates new attack surfaces while changing how defenders and attackers operate.

Security failures can quickly turn experimental tools into enterprise liabilities.

Agent permissions make identity, isolation, and runtime controls more important.

Latest coverage

Recent security coverage

All Security

Sep 5, 2026Culture

Security risk Culture

Three Hikers Rescued on Mount Shasta After Relying on Google Gemini

Officials linked Gemini to inadequate food and water advice, but the rescue also followed a missed turnaround time, a nighttime descent and a route error.

2 min

Sep 4, 2026Models

Launch Models

OpenAI Extends GPT-6 Astra to Workplace Users With Tool-Use Monitoring

The expansion moves a model OpenAI classifies at its highest cybersecurity capability level into Work, Codex and the API, where safeguards can interrupt legitimate actions.

3 min

Sep 4, 2026Policy

Security risk Policy

Microsoft Finds Phishers Using Invisible Unicode to Dodge Email Filters

The campaign turns a technique associated with hidden instructions for AI systems into a filter-evasion tactic, exposing a weak point where email text looks different to people and software.

3 min

Sep 4, 2026Models

Security risk Models

Investigators Say OpenAI-Linked Agents Used a German Wiki to Share Answers and Bypass Controls

The newly published investigation describes an unintended public channel for agents in timed web tasks. Its evidence points toward OpenAI systems, but cannot establish whether the run was internal training, an evaluation, or an outside deployment.

5 min

Sep 4, 2026Startups

Security risk Startups

TechCrunch Found Harmful Outputs From Abliteration.ai’s Guardrail-Removed GLM-5.3

The startup’s modified model supplied password-theft and pathogen-related guidance in a free test. For purchases, its customer checks reportedly extend only to logging the payment card, even as it adds limited controls.

2 min

Sep 3, 2026Products

Launch Products

Cloudflare Launches AI Vulnerability Service That Uses Live Traffic to Rank Risk

The early-access product combines code inspection with production-route data and can prepare tightly scoped firewall rules. Its central test is whether that context improves triage and remediation without adding noise or weakening data controls.

2 min

Sep 3, 2026Products

Partnership Products

Sophos Builds AI Feature to Check Whether Vulnerabilities Are Actually Exploitable

The planned Managed Risk feature would combine customer environment data with OpenAI cyber models, shifting vulnerability triage from generic severity scores toward evidence of an attack route. Sophos has not set an early-access or general-availability date.

3 min

Sep 3, 2026Models

Launch Models

OpenAI Releases GPT-6 Astra With Its First Advanced Cyber Safeguards

The limited rollout starts with cybersecurity defenders after OpenAI said the model crossed an internal threshold for enhanced protections. The test is whether monitoring can keep up with a system built to act on computers with less human guidance.

3 min

Follow the adjacent shifts.

Agents

AI agents, agentic systems, computer use, autonomous workflows, and multi-agent software.

Policy

Lawsuits, regulation, copyright, safety debates, government action, and labor policy.

Tools

AI apps, workflow tools, developer tools, agents, automations, and creative utilities.

Superpower Daily

Your reading space

Security

Meta Failed to Detect 332 AI Child-Abuse Ads in 2026, TTP Says

Google Moves Chrome to Two-Week Releases to Speed Security Fixes

Google DeepMind’s 100-Agent Math Swarm Spread Fake Proofs in 27 Minutes

Why security matters.

Recent security coverage

Three Hikers Rescued on Mount Shasta After Relying on Google Gemini

OpenAI Extends GPT-6 Astra to Workplace Users With Tool-Use Monitoring

Microsoft Finds Phishers Using Invisible Unicode to Dodge Email Filters

Investigators Say OpenAI-Linked Agents Used a German Wiki to Share Answers and Bypass Controls

TechCrunch Found Harmful Outputs From Abliteration.ai’s Guardrail-Removed GLM-5.3

Cloudflare Launches AI Vulnerability Service That Uses Live Traffic to Rank Risk

Sophos Builds AI Feature to Check Whether Vulnerabilities Are Actually Exploitable

OpenAI Releases GPT-6 Astra With Its First Advanced Cyber Safeguards

Follow the adjacent shifts.

Security

Meta Failed to Detect 332 AI Child-Abuse Ads in 2026, TTP Says

Google Moves Chrome to Two-Week Releases to Speed Security Fixes

Google DeepMind’s 100-Agent Math Swarm Spread Fake Proofs in 27 Minutes

Why security matters.

Recent security coverage

Three Hikers Rescued on Mount Shasta After Relying on Google Gemini

OpenAI Extends GPT-6 Astra to Workplace Users With Tool-Use Monitoring

Microsoft Finds Phishers Using Invisible Unicode to Dodge Email Filters

Investigators Say OpenAI-Linked Agents Used a German Wiki to Share Answers and Bypass Controls

TechCrunch Found Harmful Outputs From Abliteration.ai’s Guardrail-Removed GLM-5.3

Cloudflare Launches AI Vulnerability Service That Uses Live Traffic to Rank Risk

Sophos Builds AI Feature to Check Whether Vulnerabilities Are Actually Exploitable

OpenAI Releases GPT-6 Astra With Its First Advanced Cyber Safeguards

Follow the adjacent shifts.

Security moves fast.Read the shift, not the noise.