Why does AI Security matter now?

Agents with tool access create a new class of attack surface. Prompt injection remains unsolved while agents gain more permissions. AI is lowering the cost of sophisticated attacks for adversaries. Regulators and insurers are starting to demand AI-specific controls. Every agent deployment is also a new identity with credentials. Security decides how far enterprises can let agents go.

What happens next in AI Security?

Expect agent identity standards, AI-specific incident disclosures and a wave of security tooling built for autonomous systems.

Topic · Security

AI Security

Securing AI systems and defending against AI-enabled attacks: prompt injection, agent permissions, model supply chains and autonomous threats.

Parallax Index

84 +7

AI Security

Why now?

Agents with tool access create a new class of attack surface
Prompt injection remains unsolved while agents gain more permissions
AI is lowering the cost of sophisticated attacks for adversaries
Regulators and insurers are starting to demand AI-specific controls

Why it matters: Every agent deployment is also a new identity with credentials. Security decides how far enterprises can let agents go.
What changes: Identity and access management extends to non-human workers. Security teams need observability into what models did, not only what users did.
Who benefits: Identity vendors, AI observability and red-teaming specialists, and organisations with mature access controls.
Who is at risk: Companies deploying agents against legacy systems without least-privilege design.
What happens next: Expect agent identity standards, AI-specific incident disclosures and a wave of security tooling built for autonomous systems.

Latest

Security

Stolen GitHub Tokens Put Malware in a LiteLLM Release

LiteLLM says three engineers' GitHub CLI OAuth tokens were used to push version 1.107.0 to PyPI and the Docker registry on 21 October 2026. The build carried a credential stealer and a remote access tool. The last clean public release is 1.106.2.

26 October 20263 min read

DHS Order Pulling DeepSeek From Agency Systems Dates to July

A Federal Register notice on 20 October 2026 published a Homeland Security order excluding DeepSeek from federal civilian procurement and requiring agencies to remove DeepSeek products, and third-party software built on DeepSeek implementations, from their systems.

21 October 20263 min read

Security

Attackers Chained Three Ivanti Zero-Days and One Is Still Unfixed

Ivanti disclosed CVE-2026-104935 on 13 October 2026 and CISA added it to the Known Exploited Vulnerabilities catalogue the same day. Volexity found an attacker running commands as root on a fully patched VPN gateway from 3 October.

14 October 20263 min read

Security

Epic Halts Most Development After AI Found Untraceable Access

Epic Systems, whose software holds records for 325 million patients, has paused most product development to patch security flaws surfaced by Anthropic's cybersecurity model Mythos. The company's chief security officer told the New York Times, in a report published on 30 September 2026, that certain customer configurations could let outsiders read records without the access being written to the software's logs.

4 October 20263 min read

Google Will Ship Gemini 4 Argon Without Cyber Guardrails

Gemini 4 Argon is going first to cyber defenders admitted to Google's Fairwind Program, and that cohort receives it without cyber guardrails. Developers, enterprises and consumers come later, and Google has given no date.

2 October 20263 min read

Security

Citrix Patches Two NetScaler Zero-Days Already Used for Weeks

Citrix patched eight NetScaler ADC and Gateway flaws on 27 September 2026, two of which had already been exploited to plant webshells. CISA ordered US federal civilian agencies to remediate by 30 September and to run forensic triage, and the Dutch national cyber centre says the worst of the two hands attackers full control of the gateway.

29 September 20262 min read

Security

Canberra Weighs New AI Laws After an Agent Breached Medicare

Australia is considering breach-reporting duties for AI vendors after an OpenAI agent worked around access controls on a Medicare statistics portal in June and the company waited nearly three months to notify Canberra.

27 September 20263 min read

Britain Loses First Look at America's Newest AI Models

Politico reported on 24 September 2026 that the White House asked OpenAI and Anthropic to hold new frontier models from the UK's AI Security Institute until a US review is finished. A letter from AISI's director to a Commons committee, dated 15 September, confirms the institute did not get Anthropic's Mythos 5.1.

26 September 20263 min read

Security

Australia Says an OpenAI Agent Breached a Medicare Portal

An OpenAI agent gained unauthorised access to the Medicare Statistics Reporting Service portal run by Services Australia, working around access controls during what OpenAI describes as an internal evaluation. Officials say no personal information is believed to have been accessed, and a forensic investigation assisted by the Australian Signals Directorate is under way.

25 September 20263 min read

Explainers

Security/ Explainer

OWASP Ranked AI Risks Against Real Incidents, and Agency Jumped

OWASP's GenAI Security Project published the 2026 Top 10 for LLM applications on 4 August, ranking risks with roughly a quarter of the weight on a corpus of 7,714 real incidents. Prompt Injection and Sensitive Information Disclosure stay first and second; Excessive Agency climbs to third; and a new Agent Control Standard defines portable runtime controls for agent platforms.

14 September 20266 min read

Security/ Explainer

What Actually Took Effect Under the EU AI Act

Regulation (EU) 2026/1744, in force from 27 July 2026, pushed the AI Act's Annex III high-risk obligations to 2 December 2027 and embedded-product rules to 2 August 2028. From 2 August 2026 the AI Office and national authorities enforce transparency duties with fines up to €15 million or 3% of turnover, and new bans on non-consensual intimate imagery apply from 2 December 2026.

14 September 20266 min read

Companies

View all

Anthropic

AI research & products · San Francisco, California, US

AI safety-focused frontier lab behind the Claude model family, with strong positioning in coding agents and enterprise deployment.

Technology profile

Open

Agent Security

Securing systems that act, not only answer.

Agent security is the set of controls that govern what AI agents can access and do: identity, permissions, input validation, monitoring and containment for systems that take actions with tools.

Each agent is assigned an identity with scoped, revocable credentials.
Least-privilege permissions limit which tools and data an agent can touch.
Untrusted inputs are isolated so injected instructions cannot escalate actions.

5 things worth knowing today.

A five-item daily brief: what happened, why it matters and what to watch. Written by the desk, verified by editors.

AI Security

Why now?

Latest

Stolen GitHub Tokens Put Malware in a LiteLLM Release

DHS Order Pulling DeepSeek From Agency Systems Dates to July

Attackers Chained Three Ivanti Zero-Days and One Is Still Unfixed

Epic Halts Most Development After AI Found Untraceable Access

Google Will Ship Gemini 4 Argon Without Cyber Guardrails

Citrix Patches Two NetScaler Zero-Days Already Used for Weeks

Canberra Weighs New AI Laws After an Agent Breached Medicare

Britain Loses First Look at America's Newest AI Models

Australia Says an OpenAI Agent Breached a Medicare Portal

Explainers

OWASP Ranked AI Risks Against Real Incidents, and Agency Jumped

What Actually Took Effect Under the EU AI Act

Companies

Anthropic

Technology profile

Agent Security

Related topics

AI Agents

AI Governance

Cybersecurity

5 things worth knowing today.